Systolic architectures for artificial neural nets

A novel design of neural networks using a two-dimensional systolic array is proposed. Two techniques are applied in the design, namely, two-dimensional pipelining and multirate processing (two-level clocking). Two-dimensional pipelining operation gives significant improvement in computation time compared to the currently known 1-D and 2-D systolic implementation schemes. Multirate clocking is used so that weights (synapses) can be transmitted and passed systolically at a rate much higher than activation voltages, to achieve maximum array throughput and to eliminate global interconnections present in many arrays (including systolic) designs, thus reducing synchronization and propagation delay problems. This scheme of passing weights also saves area significantly, since local storage area for the weights can be reduced. The design is applied to the implementation of two popular neural networks namely the Hopfield neural network and the back-propagation neural network, which are supposed to be difficult to implement using the proposed scheme because of feedback and multilayer characteristics