Quickly-started chip with parallel computing cores and method for configuration of chip

Abstract

The invention provides a quickly-started chip with parallel computing cores and a method for configuration of the chip. The chip includes a plurality of computing cores and routing nodes in one-to-onecorrespondence with each computing core. One computing core and one corresponding routing node form a computing node, and a corresponding nonvolatile memory is configured for the plurality of computing cores; the nonvolatile memory stores configuration information of each computing core, and when the chip is started, each computing core obtains the configuration information from the correspondingnonvolatile memory to perform configuration. The nonvolatile memory is configured for each computing core of the chip, and the configuration information of each computing core is stored therein. Whenthe chip is powered off and restarted, each computing core in the chip can direct obtain configuration information for configuration from the nonvolatile memory without importing configuration information from the outside, and quick starting of the chip during restarting after power-off can be achieved.

Description

technical field [0001] The present invention relates to the technical field of chips, and more specifically, to a chip with a parallel computing core and a method for configuring the chip. Background technique [0002] Under the trend of big data development, parallel computing is more and more widely used. For example, the brain-like computing platform has many neurons, which is reflected in the brain-like computing platform, that is, there are many computing cores on the computing chip, and many computing cores Capable of parallel computing While many brain-inspired computing platforms have been successfully developed, many optimizations for brain-inspired computing platforms are also in progress, such as researching the feasibility of low-precision algorithms for the lack of precision of hardware systems, and researching for the volatility of hardware memory Design of non-volatile memory brain-inspired computing platform, etc. The brain-inspired computing model is becomi...

AI Technical Summary

Problems solved by technology

The brain-inspired computing model is becoming increasingly large-scale. In order to support large-scale computing models, the brain-inspired computing platform must also have large-scale computing resources. As a result, when starting the platform, it is necessary to read a large amount of configuration information from the outside to the system for internal configuration. , the time of this configuration is very long relative to the calculation, which affects the efficiency of the computing system

Images