This paper focuses on the theoretical performance of a neuro-inspired digital communication, where the input signal is below the activation threshold of the neuron, making the noise crucial for the detection. We first derive the error probability of a neuro-inspired detector build upon a single Integrate-and-Fire (IF) neuron and then extend this result to a IF-based detector build upon multiple neurons. In this case, we propose to optimize the number of neurons in parallel, leading to an error probability of below 10 −4 for a large noise range, proving the strength of such a IF-based detector.