Multitechnology neural network accelerator

The authors describe an architecture, suitable for standardization, which has the ability to be scaled well above teraflops in performance. The nature of this architecture is the combined ability to support multiple technologies such that analog and digital computing nodes can function in parallel. It features transparent nearest neighbor connectivity allowing for better than gigabaud communication rates between nodes, and modular software development allowing for system upgrades that are transparent to application writers. An example system which is being proposed for recycling of materials is discussed. It is shown how various nodes are integrated together to address a sensor fusion problem. Sample code segments are provided, illustrating how an application writer would utilize the system. The enclosure technology capable of housing the system is introduced