For miniaturized multi-gas sensors, the detected multi-gas signals would be self-interfered by responses to multiple gases. In this paper, a fast Independent Component Analysis (FICA) is proposed to restore the original source signals from the mixed signals received by different gas sensors. This FICA is designed and implemented by low power algorithm-architecture co-design considering the tradeoffs among power, delay and accuracy of extracted signals for multi-gas sensor applications. To further reduce the power consumption, a data-length controller is designed to adjust the calculated data-length. Additionally, a stability check unit is utilized to terminate the ICA execution for reduction of the computation time and total energy. Compared with the conventional ICA design, the proposed low-power FICA realizes energy reduction by 75% for multi-gas sensor applications.