This paper discusses the system architecture of a network consisting of 400 simple microprocessors for simulating neural networks with 400 nodes (neurons). In particular, the multiprocessor network should be able to simulate the CALM Neural Network approach. CALM employs a modular structure, which is used to simplify the hardware design. A system architecture is developed, which is based on (1) parallel units that consist of several processors, and (2) a special network controller that arranges the inter-unit communication.