Memristor array system capable of being used for training

Abstract

The invention belongs to the technical field of memories, and particularly relates to a memristor array system capable of being used for training. The system comprises a WL register, a BL register, an input driver, a memristor array used for storing data information, an analog-to-digital converter used for converting analog signals output by the memristor array into digital signals, and a shift adder used for processing the digital signals output by an ADC. The memristor bidirectional reading method comprises the following steps: (1) forward reading: fixing SL to be high voltage, selecting a storage unit through the voltage applied by BL, and corresponding to a forward propagation calculation process; and (2) reverse reading: changing the voltage applied to the two ends of the storage unit, fixing the BL to be high voltage, and selecting the storage unit through the voltage applied by the SL. On the premise of not changing and increasing the array unit structure, the bidirectional reading speed and margin of the resistive random access memory unit are not reduced at all, the read operation time consumption of the memristor is greatly reduced, and the bidirectional read operation is realized.

Description

Technical field [0001] The present invention belongs to the field of memory technology, specifically relates to a memristor array system that can be used for training. Background [0002] With the advent of the era of big data, the process nodes are shrinking, and the proportion of memory devices is increasing. Volatile memory, in order to keep the stored data from being lost, in the event of a power outage, it is still necessary to provide a power signal to the memory, which will lead to an additional large unit leakage power consumption. RRAM (Resistive Random Access Memory) is a new type of nonvolatile memory based on the working principle of memristems. The memristor has the advantages of simple structure, compatibility with existing CMOS processes, high microdirection, multi-value storage, easy 3D integration, etc., as a non-volatile memory, the memristor can still maintain data after power failure, so there will be a small unit leakage power consumption. [0003] A memristo...

AI Technical Summary

Problems solved by technology

[0007] In the traditional bidirectional reading method, due to the characteristics of the memristor unit itself, the loss of margin and speed will be caused, and the impact is mainly reflected in the process of reverse reading.

Images