Integration platform for multiscale modeling of neuromorphic systems

The current multilevel resistive memory elements allow increasing the integration density of nonvolatile memory as well as designing and creating systems with a parallel computing mechanism. Such devices are based on memristor elements necessary for developing the foundations of analog neuromorphic networks that are used to solve data mining problems. However, the use of memristors as a part of neuromorphic devices encounters a number of problems such as the scatter of the switching parameters (voltage and memory window) from cell to cell, asymmetry and nonlinear effects, and others. Such problems dictate the need to create original simulation models and new software tools that will allow one to evaluate the influence of disturbing factors on the predictive accuracy and network learning process. In this paper, to solve the problem of multiscale modeling of neuromorphic systems, the authors use the original information technology for constructing multiscale models. For its practical implementation, an integration platform has been built that allows one to evaluate the influence of disturbing factors on the predictive accuracy and learning process of a neuromorphic network and in the future, it will be able to provide information for a reasonable choice of materials, configuration, and topology of memory cells of new-generation computers.