A Robust Dispatching Algorithm for Autonomous Distributed Manufacturing of Mixed VLSI Products

This paper describes a new dispatching algorithm for dynamic allocation of work-in-process (WIP), named pseudo periodical priority dispatching (P3D). The algorithm is developed to manage a processing station in an autonomous distributed manufacturing system. The priority of dispatching is examined when each quantum starts, by considering both the amount of WIP in the input buffer of the processing station and the arrival rate of WIP. Here, a quantum is defined as a period, during which a single type of WIP is processed in a machine, and the machine for each type of WIP is fixed during the quantum. Performance parameters, i.e., adjustment rate, throughput, response time, and tardiness, when applying P3D are compared with the results of the first-come first-serve (FCFS) and the shortest processing time (SPT) in simulations assuming Poisson arrival. A fluctuation in arrival rate causes large inventories at the processing station. The adjustment rate is the lowest and the availability of machines is the highest with P3D; therefore, P3D can quickly dissolve a bottleneck. Both P3D and SPT produced shorter response times than FCFS in average, and both P3D and FCFS produced shorter tardiness than SPT. The small quantum size helps to shorten the response time in P3D. P3D with the minimized quantum size gives a good estimation when setting the due date. P3D achieves very fair dispatching even though the arrival rate differs by 40 times among mixed WIPs. The author concludes from the above results that P3D is a robust dispatching algorithm for manufacturing a broad mix of VLSI products.