Crossbar Neuromorphic Computing for Large Input Neuron Processing

The patent divides the neural network processing into multiple segments by grouping input neurons. Instead of requiring a single large crossbar to handle all input neurons simultaneously, the system processes neurons in groups that fit within the crossbar's axon capacity. This segmentation allows large-scale neural networks to be processed using smaller, more manageable crossbar units, reducing hardware complexity while maintaining processing capability.

The patent introduces a temporal dimension to the processing architecture by implementing sequential processing of neuron groups. Instead of expanding the crossbar area to accommodate all neurons simultaneously, the system processes neurons in time-sequential batches. Multiple groups of input neurons are processed sequentially, with each group fitting within the existing crossbar area, thereby increasing the total number of processable neurons without increasing physical crossbar size.

The patent merges multiple processing operations into a unified sequential processing framework. Instead of managing completely independent crossbar units with complex interconnections, the system combines multiple crossbars into a coordinated system where each crossbar processes specific neuron groups. The control logic unifies the operation of multiple crossbars, managing data flow and coordination to reduce overall system complexity while maintaining enhanced processing capacity.