The standard approach to parallel controller design uses sequential co
ntroller design techniques. However, since these techniques cannot rep
resent concurrent states, the problem must first be partitioned and th
en designed as a number of linked finite state machines. This initial
partitioning is usually non-optimum and limits the amount of concurren
cy in the subsequent design, where the interaction between the finite
state machines also makes verification difficult. This paper presents
an alternative technique for parallel controller design in which a syn
chronous, interpreted Petri net is used to model the controller's spec
ification as a single parallel network. No pre-partitioning is necessa
ry, and the amount of concurrency dynamically reflects the amount of p
arallel activity on the data path. The formalism provided by the techn
ique also reduces the likelihood of parallel synchronization errors. T
he technique is illustrated with the design of a parallel controller f
or a transputer link adapter and its implementation on a programmable
logic device.