Method for mapping symbolic numbers on memristor array

Abstract

The invention discloses a method for mapping symbolic numbers on a memristor array, and the method comprises the steps: directly mapping the symbolic numbers on the memristor array in a complement expression form, and obtaining a mapping scheme according to the bit widths of different mapping numbers and the precision of memristors. Firstly, whether a current memristor type device is binary or multi-valued needs to be confirmed, and if the current memristor type device is a multi-valued device, the unit precision needs to be confirmed; then, the bit width of the mapped symbolic numbers is determined, and the symbolic numbers are converted into binary numbers in a complement form; and finally, a symbolic number mapping scheme is acquired. The method is suitable for neural network calculation, consumption of memristor resources occupied by the mapped symbolic numbers of the method is small, universality is high, and the numerical value coverage range is the same as the actual expression range.

Description

technical field [0001] The invention relates to the field of novel intelligent computing, in particular to a symbol number mapping method on a memristor array. Background technique [0002] At present, research on memristors is becoming more and more extensive, involving various fields, such as security, mathematics, and artificial intelligence. In the field of artificial intelligence, memristors are often used as tools and media for neural network computing, and their characteristics of high read / write speed, low power consumption, and high parallelism are fully reflected. Memristor, Memristor (Memristor) is a circuit device that represents the relationship between magnetic flux and charge. Unlike resistors, the resistance value of memristor is determined by the charge flowing through it. By measuring the resistance value of memristor, we can know The amount of charge flowing through it has the effect of memory charge. Moreover, the integration, power consumption, and rea...

AI Technical Summary

Problems solved by technology

[0005] However, in the process of matrix calculation through memristors, for the mapping of neural network weight values, on the one hand, in the prior art, signed numbers and unsigned values ​​are mapped to a positive memristor array and a negative memristor array. On the memristor array, and then calculate the difference to get the final value, but this method will occupy more resources of the memristor, and will also lose a minimum negative value in the numerical expression range

On the other hand, the mapping to the number of symbols is realized by changing the device or adding an analog circuit. This method is more complicated and reduces the versatility

Images