The theorem of fuzzy logic was presented from Zadeh at the earliest. The main concept of the fuzzy set proposed by Zadeh is to convert the experience rule of human being into the fuzzy set of the fuzzy inference and quantify its membership by mathematical. We proposed a high-speed fuzzy inference processor suitable for the trapezoid-shaped membership function in this paper. The most outstanding characteristic of the proposed fuzzy inference processor is that we analyze the relationship between input variables and fuzzy rules first. So that we can find the membership functions that overlaps the input variables as soon as possible. And ignore the fuzzy rules those are not contributed to the result of the fuzzy inference. We can slash inference time by a wide margin because there are only fuzzy rules composed of the membership functions that overlap the input variable can we use to proceeding fuzzy inference. We use 0.35μm standard cell library presented by CIC to implement the fuzzy inference processor. We can know its working frequency reaches to 190MHz and its inference velocity is up to 23.7MFLIPS (Mega Fuzzy Logic Inferences Per Second) through the result of timing analysis. Compared with other architecture, the suitability and inference velocity of the architecture proposed in this paper exhibit quite well.