A conflict-free architecture for multi-butterfly parallel processing in-place radix-r fast Fourier transform (FFT) is proposed. The parallel FFT computing principle for any radix-r is deduced, and the number of data points and butterfly units could be the any integer power of r. The method works for FFT data paths with pipeline data stream and multiple dual-port memory banks for parallel access. By dividing the data needed to be processed into multiple independent sets, a method is given to calculate the discrete Fourier transform (DFT) with multiple parallel small points DFT. With the help of time division multiplexing radix-r butterfly, the multi-butterfly could access the memory banks without conflicts and additional hardware cost compared with the traditional algorithm. The hardware implementation block diagram of the new method are also presented, and the comparison of computing time with the traditional methods shows the effectiveness of the new architecture.