Based on the analysis of collective activities of ant colonies, the typical example of swarm intelligence, a new approach to construct swarm intelligence based multi-agent-system (SMAS)for dynamic real-time scheduling for semiconductor wafer fab is proposed. The relevant algorithm. pheromone-based dynamic real-time scheduling algorithm(PBDR), is given. MIMAC test bed data set mini-fab is used to compare PBDR with FIFO (first in first out), SRPT(shortest remaining processing time) and CR (critical ratio)under three different release rules, i.e. deterministic rule, Poisson rule and CONWIP(constant WIP). It is shown that PBDR is prior to FIFO. SRPT and CR with better performance of cycle time, throughput, and on-time delivery, especially for on-time delivery performance.