The canonical anticommutation relations (CAR) for a fermionic degree of freedom can be written as follows:
$$ a^2 = \left( a^{\dagger} \right) ^2 = 0, $$ $$ a a^{\dagger} + a^{\dagger} a = 1. $$
This admits a 2-dimensional matrix representation:
$$ a=\left(\begin{array}{cc} 0 & 1\\ 0 & 0 \end{array}\right),\;a^{\dagger}=\left(\begin{array}{cc} 0 & 0\\ 1 & 0 \end{array}\right). $$
It is easy to verify that the CAR algebra holds. The dimensionality of this representation is 2, which shouldn't come as a surprise, because due to the Pauli exclusion principle the number of fermionic excitations occupying a degree of freedom could either be 0 or 1.
Now, consider a system of $N$ fermionic degrees of freedom. The CAR algebra reads:
$$ a_i a_j = - a_j a_i, \; a^{\dagger}_i a^{\dagger}_j = -a^{\dagger}_j a^{\dagger}_i $$ $$ a_i a^{\dagger}_j + a^{\dagger}_j a_i = \delta_{ij}. $$
I am interested in the generic formula for a matrix representation of such CAR. According to the Pauli principle, it should have the dimensionality of $2^N$.
Note that simply taking
$$ a_i = 1 \otimes \dots \otimes a \otimes \dots \otimes 1 $$
gives you a different algebra – for example, $a_i$ and $a_j$ commute for $i \neq j$ instead of anti-commuting. Hence, this isn't the sought matrix representation.