Analog domain CIM (computing in memory) array structure based on magnetic random access memory

Abstract

The invention discloses an analog domain CIM (computing in memory) array structure based on a magnetic random access memory (MRAM). The analog domain CIM (computing in memory) array structure comprises a 6-transistor 2-magnetic tunnel junction (6T2M) storage array, a read-write circuit, a row decoding driving circuit, a data input unit, a pulse generation circuit, a current mirror integration module, an analog-to-digital (A / D) converter, a shift addition circuit, a time sequence control circuit and a mode selection module. The method has a standard read-write mode and an in-memory calculation mode. The read-write operation of data in the storage array is realized in the standard read-write mode; in the in-memory calculation mode, a 6T2M storage unit is utilized, so that the equivalent magnetic resistance ratio (TMR) of the MRAM is improved; multi-bit multiplication accumulation in neural network calculation is completed through current integration while data is read, and meanwhile, a calculation module and a storage array are integrated together, so that memory access energy consumption is reduced; and compared with a traditional Von Noremann architecture neural network accelerator, the calculation precision and the circuit energy efficiency are effectively improved.

Description

technical field [0001] The invention belongs to the field of integrated circuit design, and in particular relates to a magnetic random access memory (MRAM)-based analog domain in-memory calculation array structure and a circuit design method for realizing multi-bit multiply-accumulate calculation based on the memory. Background technique [0002] In recent years, the development of artificial intelligence and machine learning has brought a large amount of data and calculations, which put forward higher requirements for the throughput and speed of computing units and memories. In the traditional von Neumann architecture, the memory and the computing unit are two independent parts. The traditional digital accelerator can only accelerate the computing unit, which will cause the speed mismatch between the computing unit and the memory. At the same time, a large amount of data generated in the calculation requires a large area of ​​memory to store, and the increase in the area of...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a magnetic random access memory-based analog domain in-memory computing array structure to solve the technical problems of MRAM-based analog domain in-memory computing with low TMR and analog in-memory computing non-linearity

Images