Artificial intelligence processing device and weight parameter read-write method and device

Abstract

The invention relates to an artificial intelligence processing device and a weight parameter read-write method and device. The artificial intelligence processing device comprises a processing module and a weight storage module, the weight storage module is used for storing weight parameters, and the processing module is used for: obtaining first response resistance of a weight memory when a target neural network is trained in the nth round, determining a first weight parameter corresponding to the first response resistance according to a preset first corresponding relation between the response resistance and the weight parameter; training a target neural network according to the first weight parameter and pre-acquired training data, acquiring a second weight parameter of the target neural network of the nth round of training, acquiring a second response resistance corresponding to the second weight parameter according to the first corresponding relationship, and determining a write-in voltage corresponding to the second response resistor according to a preset second corresponding relationship; and adjusting the response resistance of the weight memory according to the write-in voltage. According to the device and the method disclosed by the embodiment of the invention, the storage performance of the weight memory can be improved.

Description

technical field [0001] The present disclosure relates to the field of computer technology, and in particular to an artificial intelligence processing device, a method and a device for reading and writing weight parameters. Background technique [0002] With the development of artificial intelligence, neural networks have been widely used in many fields, and neural networks that simulate the human brain can achieve complex recognition tasks. The training of the neural network needs to continuously read the parameters in the neural network and update them continuously until the recognition rate of the neural network reaches the required value. In order to increase the speed of accessing data during the neural network training process, the parameters of the neural network can be stored in corresponding memory. [0003] The existing data memory is generally an off-chip memory, that is, data transmission is performed between the memory and the computing unit through a bus. This ...

AI Technical Summary

Problems solved by technology

[0003] The existing data memory is generally an off-chip memory, that is, the data transmission between the memory and the computing unit is carried out through the bus. This method consumes a lot of time and energy during data transmission.

In addition, the existing data memory is often based on the analog circuit to realize data storage and data operation, in the process of data processing, the device has a large uncertainty, and when multiple storage units process data at the same time, the uncertain The performance problem is more serious; when data movement is required, it needs to consume a lot of energy and is less flexible

Images