Towards Building a Neuromorphic Computer. A Reconfigurable Hardware Continuous Time Recurrent Neural Network

      Аннотация: In this work, a complete description of the design for implementing a neuromorphic computer has been provided with details of the components and concepts used to make it feasible in hardware. The final hardware implemented, is tested with well-defined and well-analyzed configurations and the hardware behaves in concurrence with its simulated counterparts and in accordance with the differential equation guiding the underlying Neuromorhic logic (I.E. of a Continuous Time Recurrent Network (CTRNN)).The neurons and supporting devices are all constructed from commonly available off-the-shelf components and the implementation has been tested to insure a high degree of fidelity between behavior predicted by CTRNN state equations and the correspondingly programmed computer. In addition, it is demonstrated that the constructed device is easy to use and can be employed in a wide variety of research efforts without the need to develop sophisticated electrical engineering knowledge.

Ключевые слова: analogue computers, continuous time systems, neural reconfigurable architectures, recurrent neural nets, neuromorphic computer