Embedded processor, instruction set and data processing method of an integrated storage and calculation chip

Abstract

The invention discloses an embedded processor, an instruction set and a data processing method of an integrated storage and calculation chip. The method is as follows: 1) the embedded processor directly reads data from the input data cache through the data input interface; 2) when the storage-computing integrated array performs artificial neural network calculation, the data to be processed is sent to the storage-computing integrated array / logic , control the DAC to convert the data into an analog signal and perform matrix multiplication and addition operation, and then control the ADC to convert the operation result into a digital signal; 3) The embedded processor reads back the operation result and performs operations on it or activates the operation on the intermediate data cache, and then judge whether it is the last layer of neural network operation; 4) if it is not the last layer, then loop through step 2) and step 3); if it is the last layer, then perform step 5); 5) embedded processor Perform final result processing, judge the artificial neural network recognition result, and determine whether to output the result according to the recognition result.

Description

technical field [0001] The invention relates to an integrated storage and calculation chip, in particular to an embedded processor, an instruction set and a data processing method of the integrated storage and calculation chip. Background technique [0002] With traditional von Neumann computing architectures, the processor and memory are separate. When this computing architecture is applied to artificial neural network operations, a large amount of data access and computing operations are required, and the computing power consumption is high, the time is long, and the efficiency is low, which can no longer meet the applications of most scenarios, especially low power consumption. , low-latency edge computing application scenarios. [0003] In order to break through this bottleneck, the chip based on the integrated computing architecture of storage and calculation has been developed rapidly. The basic idea is to directly use the storage unit to perform multiplication and ad...

AI Technical Summary

Problems solved by technology

In the storage-computing integrated chip, the control efficiency of the general-purpose embedded central processing unit (CPU) or microcontroller (MCU) designed for the characteristics of the storage-computing integrated chip is low, and some calculations related to artificial neural networks (such as various Activation function conversion, pooling operations in convolutional neural networks, etc.) require additional circuit logic to achieve, and there is no control logic circuit that matches the core computing array / logic of the integrated storage and calculation, which is not conducive to the rapid development of integrated storage and calculation chips and integration

Images