Моделирование нейроморфных систем на переключательном уровне

Abstract

The paper gives an overview of the major hardware and software implementation models neuromorphic chips. Design framework for neuromorphic architectures in the nano-CMOS era was presented. Our approach to the design of spiking neurons and STDP learning circuits relies on parallel computational structures where neurons are abstracted as digital arithmetic logic units and communication processors. Model as a biological neural systems and formal neural systems were presented. Known model Hodgkin- Huxley and its modifications were considered. Examples of calculation model Morris-Lecar, Hindmarsh-Rose and Fitzhugh-Nagumo were performed in the system Math 9. Cellular neural networks, and a specific example L. Chua were considered. Hardware implementation of formal neural systems (threshold logic) was presented. MOS implementation of neuronal adaptation is not very biologically plausible, but simulations of adaptation suggest other possible mechanisms that may be easily implemented in a other neuromorphic circuit. Basic threshold functions (definition and classification, basic properties and universal implementation) were presented. Тhe possibility of a multi-valued logic simulation on the switching threshold level for logic circuits has been considered. Theoretical and practical aspects of the method were given. A specific example of the application of the switch-level simulation was given for threshold logic circuits. The results, in addition to the theory of the creation of neuromorphic systems may be used in the technical logical identification and diagnosis systems.