Certain types of neurons, called "grid cells", have been shown to fire on a triangular grid when an animal is navigating on a two-dimensional environment, whereas recent studies suggest that the face-centred-cubic (FCC) lattice is the good candidate for the same phenomenon in three dimensions. The goal of this paper is to give new evidences of these phenomena by considering a infinite set of independent neurons (a module) with Poisson statistics and periodic spread out Gaussian tuning curves. This question of the existence of an optimal grid is transformed into a maximization problem among all possible unit density lattices for a Fisher Information which measures the accuracy of grid-cells representations in $\mathbb{R}^d$. This Fisher Information has translated lattice theta functions as building blocks. We first derive asymptotic and numerical results showing the (non-)maximality of the triangular lattice with respect to the Gaussian parameter and the size of the firing field. In a particular case where the size of the firing fields and the lattice spacing match with experiments, we have numerically checked that it is possible to find a value for the Gaussian parameter above which the triangular lattice is always optimal. In the case of a radially symmetric distribution of firing locations, we also characterize all the lattices that are critical points for the Fisher Information at fixed scales belonging to an open interval (we call these lattices "volume stationary"). It allows us to compare the Fisher Information of a finite number of lattices in dimension 2 and 3 and to give another evidences of the optimality of the triangular and FCC lattices.