Neural network processor package

Abstract

A neural network processor (100) package comprises a plurality of storage units (100aa-100mn), and each storage unit (100ij) comprises at least one three-dimensional longitudinal memory (3D-MV) array(170) and a neural computing circuit (180). The neural network processor package (100) further comprises a first chip (100a) and a second chip (100b). The first chip (100a) includes a memory array (170), and the second chip (100b) includes a neural computing circuit (180). The memory array (170) and the neural computing circuit (180) are electrically coupled by a plurality of inter-chip connections (160).

Description

technical field [0001] The present invention relates to the field of integrated circuits, and more specifically, to a neuro-processor used by artificial intelligence (AI). Background technique [0002] A fifth application of processors is neural networks. Neural networks provide a powerful artificial intelligence tool. Figure 1A is an example of a neural network. It contains an input layer 32, a hidden layer 34 and an output layer 36. The input layer 32 contains i neurons 33 whose input data x 1 ,…x i Make up the input vector 30x. The output layer 36 contains k neurons 37, which output data y 1 , y 2 ,…y k The output vector 30y is formed. The hidden layer 34 is between the input layer 32 and the output layer 36 . It contains j neurons 35 , each neuron 35 being electrically coupled to a first neuron in the input layer 32 and a second neuron in the output layer 36 . The coupling strength between neurons is given by the synaptic weight w ij and w jk express. [00...

AI Technical Summary

Problems solved by technology

Again, the design focus of the neural network accelerator 60 is tilted toward storage—in each core, eDRAM 40 occupies 80% of the area, while NPU 30 occupies less than 10%, so the computing density is greatly limited

Images