• A
  • A
  • A
  • ABC
  • ABC
  • ABC
  • А
  • А
  • А
  • А
  • А
Regular version of the site
For visually-impairedUser profile (HSE staff only)SearchMenu

Development of a hybrid model for the design and simulation of networks-on-chip

Priority areas of development: engineering science
2020

Purpose of work

Increase in throughput capacity and reduction of hardware costs, as well as reduction of development time for new network-on-chip (NoC) structures by developing a hybrid NoC model and NoC end-to-end design principle based on it which will ensure the entire development process from the specification of general requirements of NoCs to prototype and the final description in HDL ready for implementation in ASIC in a single universal software environment.

Methodology

The object of research is NoCs.

Тhe subject of research is automation of design process of NoCs.

Empirical base of research

Empirical base of research is based on the results of many years of development in the field of NoCs design as a result of which the research team of Educational laboratory of CAD systems has its own development or deep refinement with various models of different levels of abstraction, topology generation tools, model verification tools, various specialized routing and traffic management algorithms, tools for debugging and modeling routing algorithms, tools for evaluating hardware resources for the implementation of NoCs, lightweight soft-processor cores and their programming tools, tools for prototyping and placing NoCs on one chip and even on several chips, tools for automating the simulation process, software for generating configurations of model experiments, storage, comparison and visualization of simulation results.

As part of the current research, the existing developments were supplemented and improved which made it possible to represent the development of a qualitatively new methodology for end‑to‑end design of NoCs.

Results of research

Results of research are that within the framework of research project, a search and analysis of high-level NoC models was carried out, more than 100 different models were studied and classified. A review of the most famous high-level NoC models showed that there are a huge number of solutions and approaches. Moreover, all models mainly solve particular problems, there are no comprehensive and universal models, all of them are not standardized and are often incompatible with each other. In order to conduct a NoC simulation with a combination of topology parameters, routing, traffic control and generation method, arbitration, etc., it is necessary to create an original model or modify the existing one. At the same time, it was shown that there are practically no tools for automating modeling that would facilitate the processing of simulation results and their verification, multiple launch of the model with different parameters, etc. There are no means of integrating the models with each other. This state of the problem area determines the need to develop such tools that would allow solving the indicated problems.

As part of the research project, one of the most promising high-level simulators NoC Noxim was applied and modified which significantly expanded its functionality and simulated new NoC topologies and routing algorithms. As a result of modeling, it was shown that the phenomenon of cyclic dependencies is observed in circulants which lead to the occurrence of deadlocks. To solve them, various approaches to traffic control were implemented and as a result of the simulation, the effectiveness of the method proposed by the authors in other studies was demonstrated. Also, as a result of a series of model experiments, it was shown that circulant topologies have better performance compared to the classical regular mesh and torus topologies, especially in cases where the number of nodes in the network is not a square of the number. The influence of the choice of various routing algorithms on the throughput capacity of NoCs was demonstrated the best among which turned out to be the Pair exchange algorithm.

The resulting modified model acquired a universal interface which made it possible to combine it with other lower and higher-level models to perform joint modeling and implement the principle of end-to-end design and NoC verification. At the same time, the modified Noxim model itself made it possible to neutralize the shortcomings that other models developed earlier.

Modification of universal high-level UOCNS model into the server version (UOCNS‑SE) was carried out which provided a new client-server interaction format for the model components, and the code was deeply reworked. As part of the model modification, multithreading support was introduced, and the Spring Framework was implemented which increased the speed of the model by 1.5 times, and also provided the ability to simulate simultaneously in several threads. To interact with the new model, a web interface was developed that allows modeling and aggregating its results. The documentation on the components and on the use of the simulator was expanded and supplemented. The simulator was tested, and the correctness of its work was confirmed. This high-level model is the first part of NoC hybrid model created as part of the work.

A multi-core computing system based on the schoolMIPS lightweight soft-processor cores was designed. A transceiver for connecting the computing core and the router in the network was developed. With its versatility ensured it can be used with other devices. A module for automatic generation of NoCs for various topologies was created. Complete testing of the created modules was carried out. The project was tested on a development board and it was shown that the project can be scaled for FPGA of larger capacity, as well as for several FPGAs combined into a cluster. The developed HDL-model, interface modules, and soft-processor cores are the final link in a single end-to-end NoC design system from general technical design specification to the final prototype. This low-level model is the second part of NoC hybrid model created as part of the research.

The place of hybrid models in the design process of NoCs and their advantages over the standard approach were demonstrated. The idea of using hybrid models was rethought, expanded, and embodied in the formulated principle of end-to-end NoC design when the development process from specification of general NoC requirements to the prototype and the final description in HDL ready for implementation in the form of a chip occurs in a single universal software environment. Thus, NoC development tools modified within the framework of the current research, as well as the classification of NoC models carried out, supplemented the previously obtained results of research and development and, together with the created client-server architecture, formed a single universal set of software tools that implement the topological approach, hybrid models, and the principle of end-to-end design to ensure a full cycle of NoC creation.

Level of implementation, recommendations on implementation or outcomes of the implementation of the results

The results of research are of high scientific and practical importance since they form a single information and software environment for the design of NoCs from the beginning to final implementation providing all stages of NoC design or an interface for connecting (if necessary) additional tools. 

Publications:


Романов А. Ю., Сидоренко М. В., Монахова Э. Маршрутизация в сетях на кристалле с топологией трехмерный циркулянт // Информационные технологии. 2020. Т. 26. № 1. C. 22-29. doi
Прилепко П. М., Романов А. Ю., Лежнев Е. В. Модификация высокоуровневой модели NoCModel 2.0 для моделирования сетей на кристалле с циркулянтными топологиями, in: Проблемы разработки перспективных микро- и наноэлектронных систем (МЭС-2020)..: ИППМ РАН, 2020. С. 23-30. 
Monakhova E., Monakhov O., Романов А. Ю., Лежнев Е. В. Analytical Routing Algorithm for Networks-on-Chip with the Three-dimensional Circulant Topology, in: 2020 Moscow Workshop on Electronic and Networking Technologies (MWENT).: IEEE, 2020. С. 1-6. 
Romanov A. Y., Stempkovsky A. L., Lariushkin I. V., Novoselov G. E., Solovyev R. A., Starykh V. A., Romanova I. I., Telpukhov D. V., Mkrtchan I. A. Analysis of Posit and Bfloat Arithmetic of Real Numbers for Machine Learning // IEEE Access. 2021. Vol. 9. P. 82318-82324. doi