Dear Lammps users
I have a few microns of silicon wall in a lab and need to simulate it.
My running system is incapable of parsing this structure in the usual ways.
Is the CG method used to build this silicon wall ?
If yes, how should I proceed and if not, is there an alternative way?
You tell us, thats your silicon wall and you did not share any information about it.
In general, it is much more common to simulate micrometer structures using coarse-grained models, but recent atomic simulations using billions of atoms have been able to reach such scale, so both are possible.
What do you want to learn from the simulation, that makes you believe you need to simulate the entire system? Please keep in mind that even if you would sort out the system size problem, you still have to content with a time scale issue. With an atom-level resolution simulation, you cannot look at very long times (perhaps nanoseconds, microseconds only if you get a large allocation on the biggest supercomputers available). However, processes that require the micrometer length scale are typically much slower.
Can you be a bit more specific? How many atoms do you have? Did you compile LAMMPS so it can handle such a large system? Did you check out 3.5. Optional build settings — LAMMPS documentation ?
Typical coarse graining will give you a small relief in terms of system size and also time scale, but I doubt that it will be enough for the system size you are mentioning. Those are usually studied with finite element methods (i.e. a discretized continuum approach).
I think the biggest issue is to provide an answer for my first question. Only if you understand what you need to learn and what a specific type of model can provide for the computational resources available to you, can you decide how to proceed. The general idea of just setting up a simulation of a given system and then running a simulation of it and then answers will present themselves, is a misunderstanding of what simulations are good for and how you implement them.
I wish I had all that HPC time to waste 
The goal is to calculate the temperature of argon gas between two silicone walls
In this structure, the number of gas atoms is not a large number, but each wall has about 3 billion atoms
To create such a structure in the CG way, I don’t know where and what software I should use…
If you simulate 3 billion atoms of a solid wall that is not even the central focus of your simulation, I am pretty sure that you are not doing it right.
Simulating 3 billion atoms of the walls would only make sense if you are looking at intrinsically large-scale effects inside the silicon walls.
I agree with @simongravelle, doing the walls with explicit atoms is overkill. You can probably do some kind of “extreme coarse graining” in your case and use “fix wall”. For that you would need to derive parameters (e.g. Morse or Lennard-Jones that are representative for the Argon/wall interactions). A literature search might yield something useful, but - if needed - it can also be done empirically by putting a single Argon atom in front of a small chunk of wall with explicit atoms and then move it away and record the change in energy. Since you are likely using LJ interactions for the Si/Argon, it would be possible to derive the wall parameters analytically (a wall of 12-6 explicit particles becomes a 10-4 wall with a continuous potential). At that point the concern about performance should become a non-issue, since you would need to compute only interactions for the gas atoms with the walls and each other, but not for the (many) wall atoms.
Are you aware that the temperature of the atoms in an MD simulation is usually an input parameter?
So you would have to explain with more detail what you would want to compute how.