The simulation I ran often doesn't converge, always the RHS "right hand side" over 1e10. It can be I changed the length of the terminal, the carrier recombination velocity, the light intensity or something else. I am wondering if there is a general solution to this?
For convergence, the first and the most effective method is to do proper meshing. Often a uniform mesh size results in convergence failures. Meshing needs to be done in an adaptive manner. It should be coarse in regions which are not critical, while it must be fine in regions where there is an interface, an abrupt change in doping level, a high electric field region, thin films, etc. First do this, most of your problems will be solved. Next, comes the choice of the model. You can study the sentaurus manuals a bit more carefully to check if the model is applicable and you are using it in the right manner. If everything fails, then the other ways to get convergence is to go to the Math section and relax the tolerance, increase the no. of Digits.
Next comes the Solve section. If you are simulating only electrical quantities, then it is often a nice idea to solve Poisson equation, electron continuity and hole continuity in 2 or 3 steps for the initial solution. For optical simulations involving intensity and wavelength, it is very essential to define mesh sizes in a different manner. Suppose your optical wavelength in the material is lambda (this means the free space wavelength divided by the refractive index). Then try to ensure that the mesh size in a region that absorbs light is such that there are at least 10 to 20 mesh points within one lambda distance in the structure. Violating such rules can also lead to non-convergence. Hope I was of help.
I am not experienced with Sentaurus but I have used other device simulation codes that have convergence problems that might be similar to your problem. The difficulty is a problem with convergence when using the iteration methods built into the code. Solutions to this convergence problem that I have found to be useful are different depending on whether you are looking for a steady-state solution or a transient solution to a physical problem. If the physical problem to be solved is steady-state, you need to give the code the inputs that are needed to make its initial guess close enough to the final solution to get convergence. This can be a trial and error process. If the physical problem to be solved is transient and starts with an initial condition, you might have to instruct the code to start off with incredibly small time steps to get convergence at later time steps. An "Incredibly small time step" means a time step too small to be physically significant (maybe quantum mechanics does not even recognize such a short time but I am not an expert on that) but is still needed for mathematical convergence of the iteration method. This is also trial and error because the code can crash at any time so you might have to adjust the time steps used to continue the iteration. The above methods, usually done with some (sometimes a lot of) trial and error, have solved all of the convergence problems that I have encountered unless the physical problem to be solved has no solution.
For convergence, the first and the most effective method is to do proper meshing. Often a uniform mesh size results in convergence failures. Meshing needs to be done in an adaptive manner. It should be coarse in regions which are not critical, while it must be fine in regions where there is an interface, an abrupt change in doping level, a high electric field region, thin films, etc. First do this, most of your problems will be solved. Next, comes the choice of the model. You can study the sentaurus manuals a bit more carefully to check if the model is applicable and you are using it in the right manner. If everything fails, then the other ways to get convergence is to go to the Math section and relax the tolerance, increase the no. of Digits.
Next comes the Solve section. If you are simulating only electrical quantities, then it is often a nice idea to solve Poisson equation, electron continuity and hole continuity in 2 or 3 steps for the initial solution. For optical simulations involving intensity and wavelength, it is very essential to define mesh sizes in a different manner. Suppose your optical wavelength in the material is lambda (this means the free space wavelength divided by the refractive index). Then try to ensure that the mesh size in a region that absorbs light is such that there are at least 10 to 20 mesh points within one lambda distance in the structure. Violating such rules can also lead to non-convergence. Hope I was of help.
I agree to the above comments:
1. Mesh
2. Tolerance (Math)
3. Digits (Math)
4. ExtendPrecision (Math)
5. Poisson with only relevant carriers (Solve)
6. Change min/max step (Solve)
7. At least the Solver itself sometimes help
Best
I agree also with the above comments. Moreover, be aware that set a voltage to an electrode without any progressive voltage increasing must induce non convergence during the solution of your initial condition.
I don't have too much experience as I'm currently having convergence issues in my simulations (specofically those involving fast transients as SEEs or III-V semiconductors). Nevertheless, if the problem is in the mesh, I've found the CNormPrint command very usefull to locate the problematic nodes in the structure.
Another usual suggestion is to deactivate some models and enable them one at the time to seek for the one that may be generating unconvergencies. Hope this helps.
Regards
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Condensed Matter Physics
- Solid State Physics