Architectures of high-performance VLSI for custom computing systems

The article deals with the design of specialized VLSI designed for the construction of high-performance computing systems. Common options for building such systems are general-purpose processors (CPUs), graphics accelerators (GPUs) that work in the mode of general-purpose computing acceleration, and FPGAs that are configured to implement a specific task or a narrow class of tasks. If there is a demanded class of tasks with specific requirements for computing resources, it is possible to develop a specialized VLSI that is more effective than CPU and GPU in terms of the use of crystal resources, and more productive than FPGA due to the use of hardware-implemented connections instead of programmable cells and switched trace lines. Taking into account modern trends in microelectronics and existing technological lag from leading global manufacturers, an important role is played by the choice of VLSI architecture as well as the use of architectural and circuit design solutions that reduce technical risks and provide solutions to urgent problems for Russian companies.