For the matrix form of regression $Y = X\beta + \epsilon$, what is the necessary and sufficient condition for the OLS estimator $\hat{\beta} = (X^T X)^{-1} X^T Y$ to be unique?