Application features of neural network structures and their elements to increase productivity of onboard digital computing systems

Автор: Antimirov V.M., Umanskii A.B., Vdovin A.S., Grishanova I.V., Rogacheva T.V.

Журнал: Сибирский аэрокосмический журнал @vestnik-sibsau

Рубрика: Математика, механика, информатика

Статья в выпуске: 2 т.17, 2016 года.

Бесплатный доступ

The article presents the results of several generations of onboard digital computing systems produced by SPA of automatics, describes the features of applied components. It examines the basic principles of improving the accuracy of structural calculations and circuit techniques including high precision calculations at tight constraints on the scale of the data processing hardware for each generation of the basic. The redundant single-processor onboard digital computers (ODC) and the problems encountered during their implementation are considered. The article describes the first ODC architecture and how they differ from those that existed at the time of computers for general use. The author emphasizes attention to logic elements created by SPA of automatics for digital blocks. The structure of the redundant ODC is analyzed. The problems associated with the choice of the structure are described. The authors list the changes in the architecture of ODC that occurred in the second generation. The authors examine “Malakhit” family ODC which implements the principle of modular construction systems, their architectural and structural differences from the ODC of the past generation. These differences cause the superiority of “Malakhit” ODC parameters as power consumption, weight, dimensions, performance and informative capacity. The authors refer to Malakhit-2 ODC as an attempt to create a high-speed system. The article describes magistral and modular onboard digital computer system (ODCS), the purpose and principle of their interaction module. The authors offer the application of a matrix neural calculator for solving the transformation of coordinate systems. The principles of its operation and the resulting gain in speed of solving the problem are examined. The authors touche upon the issue of the transition from large-scale integrated circuits with software control to the extra-large integrated circuits (VLSI), implementing neural network computation, and the creation of a new generation of Malakhit -7 ODC. The basic directions for further work on the creation of computer systems are stated. The authors consider such promising areas as the creation of distributed systems and the creation of on-chip system in the form of VLSI. In conclusion, it is noted that the use of neural network structures can be seen as one way to significantly improve the performance of digital computer systems.

Еще

Control system, digital computer, control unit, neuroprocessor, on-chip system

Короткий адрес: https://sciup.org/148177558

IDR: 148177558

Статья научная