I am one of the Featuretools developers. calculate_feature_matrixcurrently uses only functions from Pandas / Numpy / Scipy to improve performance compared to raw Python. There are several areas where using numba or Cython can help, especially in the class PandasBackendand in the functions of computing individual functions.
However, this requires a C compiler or compiled C code, and thus adds additional complexity to the installation. Due to this complexity, it is currently not very high on our list of priorities, but we may consider adding it in the future.
Instead, we focus more on scalability to larger data arrays, which is associated with parallelization, and not with the optimization of routines.