A spike derivative based feature extraction algorithm is reported in this paper. The theoretical framework includes neuronal geometry signatures and noise shaping. Through evaluating neuronal geometry signatures with compartment model, we obtain improved differentiation among similar spikes by emphasizing high frequency signal spectrum. Through studying the noise properties, we find the total noise reduces by taking the derivative of spikes, which is the simplest frequency shaping filter boosting the high frequency signal spectrum. In addition, a preliminary hardware implementation to extract spike features has been realized using an integrated microchip interfaced with a personal computer.