How do I call the ". fatbin" file through "cuda c"?

I use the “nvcc” command to compile the kernel side code into a fatbin file, so how to call fatbin through cuda c to run the kernel.

You would use the driver API. The driver API has various module load functions that can accept a fatbin, including cuLinkAddData when device-linking is required, and cuModuleLoadData or cuModuleLoadDataEx when it isn’t. You can find these functions described here in the module load functions for the driver API, with indications of usability with fatbin. I won’t be able to give a complete tutorial here on driver API usage, but you can find many worked examples (such as here) and also CUDA sample codes that demonstrate the use of the driver API to load modules such as vectorAddDrv.