Integrated circuit chip device

Patent metadata

Field	Value
Publication number	US-11748605-B2
Application number	US-202017134487-A
Country	US
Kind code	B2
Filing date	Dec 27, 2020
Priority date	Dec 27, 2017
Publication date	Sep 5, 2023
Grant date	Sep 5, 2023

Abstract

An integrated circuit chip device and related products are provided. The integrated circuit chip device is used for performing a multiplication operation, a convolution operation or a training operation of a neural network. The device has the advantages of small calculation amount and low power consumption.

First claim

The invention claimed is:

1. An integrated circuit chip device, comprising:
a main processing circuit; and
a plurality of basic processing circuits;
wherein:
the plurality of basic processing circuits are arranged as an array having h rows and w columns, each basic processing circuit is connected to an adjacent basic processing circuit, the main processing circuit is connected to w basic processing circuits in a first row, w basic processing circuits in an h th row, and h basic processing circuits in a first column;
the main processing circuit is configured to:
perform respective neural network computations in series; and
transfer data to the basic processing circuits that are connected to the main processing circuit;

at least one of the plurality of basic processing circuits is configured to:
perform computations in the neural network in parallel according to the transferred data; and
transfer computation results to the main processing circuit through the basic processing circuits that are connected to the main processing circuit.

2. The integrated circuit chip device of claim 1 , wherein:
the main processing circuit is configured to:
obtain a data block to be computed and a computation instruction, wherein,
when the computation instruction is for multiplying a matrix S by a matrix P, the main processing circuit is configured to:
transfer data in some or all rows of the matrix S to the h basic processing circuits in the first column through horizontal data input interfaces using a control circuit; and
transfer data of some or all columns of the matrix P to the w basic processing circuits in the first row and the w basic processing circuits in the h th row through vertical data input interfaces.

3. The integrated circuit chip device of claim 2 , wherein:
the control circuit of the main processing circuit is configured to:
transfer a number or some numbers of data in a plurality of columns of the matrix P to a basic processing circuit.

4. The integrated circuit chip device of claim 2 , wherein:
the at least one of the plurality of basic processing circuits is configured to:
after receiving the data of the matrix S, transfer the data of the matrix S to a subsequent basic processing circuit connected to the basic processing circuit through a horizontal data output interface of the basic processing circuit; and
after receiving the data of the matrix P, transfer the data of the matrix P to a subsequent basic processing circuit connected to the basic processing circuit through a vertical data output interface of the basic processing circuit.

5. The integrated circuit chip device of claim 1 , wherein:
the at least one of the plurality of basic processing circuits is configured to:
perform multiplication on one group of two sets of data or a plurality of groups of two sets of data at a time;
accumulate results in a register and/or an on-chip cache;
compute an inner product of one group of two vectors or a plurality of groups of two vectors at a time;
accumulate results in the register or on-chip cache; and
after obtaining a result by computing, output the result through a data output interface.

6. The integrated circuit chip device of claim 1 , wherein:
the at least one of the plurality of basic processing circuits is configured to:
when the basic processing circuit has an output interface that is directly connected to the main processing circuit, output a result through the interface;
when the basic processing circuit does not have any output interface that is directly connected to the main processing circuit, output a result towards a basic processing circuit that is capable of outputting a result to the main processing circuit directly; and
after receiving a computation result from another basic processing circuit, transfer the computation result to yet another basic processing circuit that is connected to the basic processing circuit or to the main processing circuit.

7. The integrated circuit chip device of claim 2 , wherein:
when the computation instruction is for performing a fully connected computation, the integrated circuit chip device is further configured to:
when input data of a fully connected layer is a vector, use a weight matrix of the fully connected layer as a matrix S and an input vector as a vector P; and
when the input data of the fully connected layer is a matrix, use the weight matrix of the fully connected layer as the matrix S and the input data as a matrix P.

8. The integrated circuit chip device of claim 1 , wherein:
the main processing circuit is configured to obtain a data block to be computed and a computation instruction, wherein:
when the computation instruction is for performing a convolution computation, the main processing circuit is configured to:
transfer data in some or all convolution kernels of a weight to the h basic processing circuits in the first column through horizontal data input interfaces using a control circuit; and
classify input data according to positions of convolution, wherein the control circuit of the main processing circuit is configured to:
transfer data of some or all positions of convolution in the input data to the w basic processing circuits in the first row and the w basic processing circuits in the h th row through vertical data input interfaces.

9. The integrated circuit chip device of claim 8 , wherein the control circuit of the main processing circuit is configured to:
transfer a number or some numbers of data of a convolution kernel in the weight to a basic processing circuit; or
transfer a number or some numbers of data of some convolution kernels in the weight to a basic processing circuit.

10. The integrated circuit chip device of claim 8 , wherein:
the integrated circuit chip device is configured to perform a plurality of operations of a forward operation or a backward operation, wherein:
the plurality of operations include at least one of:
a matrix-multiply-matrix computation, a matrix-multiply-vector computation, a convolution computation, or an activation computation.

11. The integrated circuit chip device of claim 1 , wherein:
the main processing circuit is configured to:
obtain a data block to be computed and a computation instruction;
classify the data block to be computed into a distribution data block and a broadcasting data block according to the computation instruction;
partition the distribution data block to obtain a plurality of basic data blocks;
distribute the plurality of basic data blocks to the basic processing circuits connected to the main processing circuit; and
broadcast the broadcasting data block to the basic processing circuits connected to the main processing circuit;

the at least one of the plurality of basic processing circuits is configured to:
perform inner product computations on the basic data blocks and the broadcasting data block to obtain computation results; and
transfer the computation results to the main processing circuit; or

the at least one of the plurality of basic processing circuits is configured to:
forward the basic data blocks and the broadcasting data block to another basic processing circuit, wherein the another basic processing circuit is configured to:
perform inner product computations to obtain computation results; and
transfer the computation results to the main processing circuit; and

the main processing circuit is configured to process the computation results to obtain an instruction result of the data block to be computed and the computation instruction.

12. The integrated circuit chip device of claim 11 , wherein:
the main processing circui

Assignees

• Cambricon Tech Corp Ltd

Inventors

• Liu Shaoli
• Song Xinkai
• Wang Bingrui
• Zhang Yao
• Hu Shuai

Classifications

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/063 Primary

  using electronic means · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title

• G06F7/5443

  Sum of products (for applications thereof, see the relevant places, e.g. G06F17/10, H03H17/00) · CPC title