6 Replies Latest reply on May 14, 2013 1:06 PM by Jared Conway

    Nitrogen purge simulation help needed

    Bjorn Sorenson
    Bjorn Sorenson May 10, 2013 12:22 PM

      I'm trying to simulate a load-lock for a semi-conductor type application.  I have a boat full of very thin wafers that will enter a small chamber which will then be purged with Nitrogen to remove the Oxygen content before the boat is moved into the main process chamber.  I'm running a time-dependent simulation to figure out how long it will take for the Oxygen in the small chamber to reach a certain ppm.  I'm fine with the mass fraction analysis part of it, but what I'm having trouble with is setting up a mesh that will solve in a reasonable amount of time.  My first attempt, I had a fairly loose mesh for the majority of the chamber, and a local initial mesh around the boat full of wafers with narrow channel refinement selected, but it ran overnight and didn't even finish the mesh calculation!  Any ideas?

       

      Thanks in advance for any insights you can provide.

        • Re: Nitrogen purge simulation help needed
          Jared Conway
          Jared Conway May 10, 2013 4:00 PM (in response to Bjorn Sorenson)

          sounds like you may have cranked up the mesh too far

          how many cells were in the coarse mesh solution?

          what values did you change for the next one? how many cells did it end up generating

           

          what i'd suggest doing is checking the "manual" settings behind the automatic settings. make sure one parameter or another isn't increasing significantly causing too many cells and move slowly upwards in mesh refinement.

           

          the next part you'll probably have to deal with is having the time dependent study run in a reasonable time. start with a coarse manual mesh setting and then a smaller manual time step that captures the overall behavior and then you can go to automatic time stepping.

           

          overall this assumes you've done everything you can to simplify the analysis (simplify components, remove components that aren't important, leverage symmetry..etc) and that there aren't any modeling problems/errors (like invalid contacts).

            • Re: Nitrogen purge simulation help needed
              Bjorn Sorenson
              Bjorn Sorenson May 13, 2013 12:02 PM (in response to Jared Conway)

              I started both the basic and the local initial mesh w/ the default value of 3, resulting in >12million cells.  I did forget symmetry, but even 6 million cells is too many for my poor computer.  No modeling problems/errors encounterd, so the simulation will run, but I need results for a customer in pretty short order.

               

              The only difference between the two meshes was turning on narrow channel refinement for the local mesh around the wafers.  I guess I can start w/ a value of 1 and a larger time step and work my way back to a more refined result, but this also sounds pretty time consuming.

               

              I'm afraid of simplifying the analysis, because the nooks and crannies between the wafer are really what's going to cause the O2 to stick around.  If I just make it a block, or even a porous block w/ geometry that doesn't accurately reflect the real situation, I can't trust my estimate of the total purge time.

               

              I guess I was hoping for a silver bullet, but I'll start w/ the coarse mesh/time step method.

               

              Thanks for the suggestions guys.

                • Re: Nitrogen purge simulation help needed
                  Bill McEachern
                  Bill McEachern May 13, 2013 12:47 PM (in response to Bjorn Sorenson)

                  What is your metric for knowing that the oxygen level is low enough? I know you mentioned that you need to be below some ppm. Is that every where, averaged over some volume,...what is it exactly? What can or do you measure to verify it?  I am not sure that the "nooks and crannies" argument is a good one as diffusion will occur regardless - yeah having a lot of connvection is a good thing but the eddies you speak of in the nooks and crannies are not stationary and even with 12 million cells would they be adequately resolved to accurately predict the behaviors you are assuming they do in the 12 million cell model? Sounds like your job is get the best answer you can with what you have. Try a simplified model with some nooks and crannies but something that will be solvable on your system. Then try the same thing with some simplifications and see if they are any different. You can still get the mass fraction in a porous media - their will be resistances. I bet it gets you an answer in a reasonable amount of resources and a answer is better than no answer especially if you can manage (bound) the uncertainties. Good luck.

                    • Re: Nitrogen purge simulation help needed
                      Bjorn Sorenson
                      Bjorn Sorenson May 14, 2013 12:17 PM (in response to Bill McEachern)

                      The spec comes from our customer's process engineers who say that a certain amount of Oxygen is detrimental to their process. We use an O2 analyzer to measure ppm, so in real life, it will be concentration at the point of the sensor, and the trick will of course be to locate the sensor in a "representative" location.  For simulation, because it's easy to let the computer do the calculation, I've got global goals for min, max, ave and bulk ave just to make post-processing of the results easy on me.  You are right, I do need some kind of answer, and none of this (even going back to the customer's requirement based on theory and possibly some experimentation) is 100% reliable.  Flow simulation has allowed us to weed out some design flaws before we build the machine, so the flow trajectories are almost as important to me as the mass fraction analysis, and this requires a fairly realistic model.

                       

                      I did reduce the local mesh around the wafers to a resolution of 1 and got a good looking mesh which reduced the total cells for the model down to about 1.3 million which is solveable.  And as predicted, I'm now running into time-step problems (an overnight solution provided almost 0.05 seconds of physical time, woo hoo!).  So my question now is:  Since I only care about the total physical time to reach steady state and the steady state quantities of O2, is there a way to get the total physical time to reach steady state by solving the non-time-dependent solution? 

                        • Re: Nitrogen purge simulation help needed
                          Jared Conway
                          Jared Conway May 14, 2013 1:06 PM (in response to Bjorn Sorenson)

                          negative ghost rider. steady state is solving a completely different set of equations, it doesn't know the time to steady state.

                           

                          your only option is to go with manual time stepping that is reasonable and remove all the periodic saving to speed up your calculation.

                           

                          i would also make sure that you don't get sensible results with a coarse mesh. in our experience, customers tend to go straight to a "good mesh" without first seeing how things work with an ok mesh. many times the results are well within reasonable tolerance in a fraction of the time.

                  • Re: Nitrogen purge simulation help needed
                    Bill McEachern
                    Bill McEachern May 11, 2013 9:36 AM (in response to Bjorn Sorenson)

                    chances are is that your "boat full of waffer"s can be idealized as a porous media and this will greatly reduce the cell count. this assumes that your mesh was well onstructed in the first place. Otherwise then Jared's comments may be a better place to start.