Abstract: A processing chip includes a first sub-circuit, an asynchronous clock interface circuit, and a second sub-circuit. The asynchronous clock interface circuit includes a first beat conversion circuit, a first trigger circuit, and a second beat conversion circuit. A first signal end of the first sub-circuit is coupled to a trigger input end of the first trigger circuit. A trigger output end of the first trigger circuit is separately coupled to an input end of the first beat conversion circuit and an input end of the second beat conversion circuit. An output end of the first beat conversion circuit is coupled to a third signal end of the first sub-circuit. An output end of the second beat conversion circuit is coupled to a second signal end of the second sub-circuit.

Abstract: A neural network on-chip mapping method and apparatus based on a tabu search algorithm are provided. The method includes: constructing a tabu search table and using a heuristic-based iterative search process to select local computing cores of a network-on-chip as candidates, establishing an integer programming model and solving an optimal solution, continuously reducing an objective cost function of a mapping solution by loop iteration, and finally obtaining an approximately optimal deployment scheme.

Abstract: In an object recognition method, an object recognition device obtains AER data of a to-be-recognized object, wherein the AER data includes a plurality of AER events of the to-be-recognized object, each AER event comprising a timestamp and address information. The object recognition device extracts a plurality of feature maps of the AER data. Each feature map including partial spatial information and partial temporal information of the to-be-recognized object, and the partial spatial information and the partial temporal information are obtained based on the timestamp and the address information of each AER event. The object recognition device then recognizes the to-be-recognized object based on the plurality of feature maps of the AER data.

Abstract: The present invention discloses a neuromorphic computer supporting billions of neurons, comprising hierarchical extended architecture and algorithmic process control within the architecture; the architecture comprises multiple neuromorphic computing chips with hierarchical organization management for implementing computing tasks, each containing computing neurons and synaptic resources and forming a neural network, spike events between computing neurons within the architecture are transmitted through a hierarchical transmission mode; the algorithmic process control comprises controlling parallel processing of computing tasks within the architecture, controlling management of synchronization time within the architecture, and controlling reconstruction of neural networks within the architecture to achieve fault tolerance and robust management of computing neurons and synaptic resources. The neuromorphic computer can support spiking neural network inference calculations with a neuron scale of billions.

Abstract: Disclosed are a parallel method and device for convolution computation and data loading of a neural network accelerator. The method needs two input feature maps and two convolution kernel cache blocks, and sequentially stores the input feature maps and 64 convolution kernels into cache sub-blocks according to a loading length, so as to execute convolution computation and simultaneously load data of a next group of 64 convolution kernels.

Abstract: In an object recognition method, an object recognition device obtains AER data of a to-be-recognized object, wherein the AER data includes a plurality of AER events of the to-be-recognized object, each AER event comprising a timestamp and address information. The object recognition device extracts a plurality of feature maps of the AER data. Each feature map including partial spatial information and partial temporal information of the to-be-recognized object, and the partial spatial information and the partial temporal information are obtained based on the timestamp and the address information of each AER event. The object recognition device then recognizes the to-be-recognized object based on the plurality of feature maps of the AER data.