# US 2024/0256828 A1 · Neural processor

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2024/0256828 A1
- Document ID: US20240256828A1
- Status: Published application
- Kind: A1
- Application number: 18/601,739
- Filed: 2024-03-11
- Published: 2024-08-01
- Earliest claimed: 2018-06-22
- Assignees: Samsung Electronics Co., Ltd.
- Applicants: Samsung Electronics Co., Ltd.
- Inventors: Ilia Ovsiannikov; Ali Shafiee Ardestani; Joseph H. Hassoun; Lei Wang; Sehwan Lee; JoonHo Song; Jun-Woo Jang; Yibing Michelle Wang; Yuecheng Li
- Web page: https://patentgrep.com/patent/US20240256828A1

## CPC

- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/063: using electronic means
- G06F 17/153: {Multidimensional correlation or convolution}
- G06F 17/16: Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)}
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/0495: Quantised networks; Sparse networks; Compressed networks
- G06N 3/08: Learning methods
- G06T 9/002: {using neural networks}
- G06F 9/3001: {Arithmetic instructions}
- Y02D 10/00: Energy efficient computing, e.g. low power processors, power management or thermal management
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- Y02D: CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE

## Abstract

A neural processor. In some embodiments, the processor includes a first tile, a second tile, a memory, and a bus. The bus may be connected to the memory, the first tile, and the second tile. The first tile may include: a first weight register, a second weight register, an activations buffer, a first multiplier, and a second multiplier. The activations buffer may be configured to include: a first queue connected to the first multiplier and a second queue connected to the second multiplier. The first queue may include a first register and a second register adjacent to the first register, the first register being an output register of the first queue. The first tile may be configured: in a first state: to multiply, in the first multiplier, a first weight by an activation from the output register of the first queue, and in a second state: to multiply, in the first multiplier, the first weight by an activation from the second register of the first queue.

## Claims

### Claim 1

A processor, comprising:
a register for storing an arrangement of weights;
a buffer for storing an arrangement of features;
a first multiplier; and
a second multiplier;
wherein the first multiplier is configured to perform a convolution of an arrangement of the features with an arrangement of weights, the performing of the convolution comprising:
forming a product of the arrangement of weights with a first portion of the arrangement of features;
forming a product of the arrangement of weights with a second portion of the arrangement of features, the second portion being offset from the first portion by n arrangement of elements in a first direction, n being a positive integer; and
forming a product of the arrangement of weights with a third portion of the arrangement of features, the third portion being offset from the second portion by one arrangement of elements in a second direction, perpendicular to the first direction,
wherein the second portion and the third portion are spaced apart from an end of a row of the arrangement of features.

### Claim 2 (depends on 1)

The processor of claim 1, wherein the performing of the convolution further comprises, based on the forming of the product of the arrangement of weights with the third portion:
forming a product of the arrangement of weights with a fourth portion of the arrangement of features, the fourth portion being offset from the third portion by m arrangement of elements in a third direction, opposite to the first direction, m being a positive integer, and
forming a product of the arrangement of weights with a fifth portion of the arrangement of features, the fifth portion being offset from the fourth portion by one arrangement of elements in the second direction.

### Claim 3 (depends on 2)

The processor of claim 2, wherein m equals n.

### Claim 4 (depends on 3)

The processor of claim 3, wherein n equals 1.

### Claim 5 (depends on 1)

The processor of claim 1, wherein the performing of the convolution further comprises based on the forming of the product of the arrangement of weights with the first portion:
forming n−1 products of the arrangement of weights with n−1 respective portions of the arrangement of features, the portion in a k-th product, of the n−1 products, being offset from the first portion by k+1 array elements in the first direction.

### Claim 6 (depends on 5)

The processor of claim 5, further comprising a cache, connected to the buffer and configured to supply features to the buffer, the cache having a size sufficient to store H+(H+n)*(W−1)−1 features, wherein:
H is a size of the arrangement of weights in the first direction, and
W is a size of the arrangement of weights in the second direction.

### Claim 7 (depends on 1)

The processor of claim 1, wherein:
the buffer is configured to include:
a first queue connected to the first multiplier, and
a second queue connected to the second multiplier,
the first queue comprises a first register and a second register adjacent to the first register, the first register being an output register of the first queue,
the first multiplier is further configured:
in a first state:
to multiply a first weight by an activation from the output register of the first queue, and
in a second state:
to multiply the first weight by an activation from the second register of the first queue.

### Claim 8 (depends on 7)

The processor of claim 7, wherein, in the second state, the output register of the first queue contains zero.

### Claim 9 (depends on 7)

The processor of claim 7, further comprising:
a first adder, configured, in the first state:
to be connected to
an output of the first multiplier, and
an output of the second multiplier, and
to add:
a product received from the output of the first multiplier, and
a product received from the output of the second multiplier.

### Claim 10 (depends on 9)

The processor of claim 9, further comprising a second adder, configured, in the second state, to be connected to the output of the first multiplier.

### Claim 11

A method for calculating with a processing circuit, the processing circuit comprising:
a register for storing an arrangement of weights;
a buffer for storing an arrangement of features;
a first multiplier; and
a second multiplier;
the method comprising performing a convolution of an arrangement of features with an arrangement of weights, the performing of the convolution comprising:
forming a product of the arrangement of weights with a first portion of the arrangement of features;
forming a product of the arrangement of weights with a second portion of the arrangement of features, the second portion being offset from the first portion by n arrangement of elements in a first direction, n being a positive integer; and
forming a product of the arrangement of weights with a third portion of the arrangement of features, the third portion being offset from the second portion by one arrangement of elements in a second direction, perpendicular to the first direction,
wherein the second portion and the third portion are spaced apart from an end of a row of the arrangement of features.

### Claim 12 (depends on 11)

The method of claim 11, wherein the performing of the convolution further comprises, based on the forming of the product of the arrangement of weights with the third portion:
forming a product of the arrangement of weights with a fourth portion of the arrangement of features, the fourth portion being offset from the third portion by m arrangement of elements in a third direction, opposite to the first direction, m being a positive integer, and
forming a product of the arrangement of weights with a fifth portion of the arrangement of features, the fifth portion being offset from the fourth portion by one arrangement of elements in the second direction.

### Claim 13 (depends on 12)

The method of claim 12, wherein m equals n.

### Claim 14 (depends on 13)

The method of claim 13, wherein n equals 1.

### Claim 15 (depends on 11)

The method of claim 11, wherein the performing of the convolution further comprises, based on the forming of the product of the arrangement of weights with the first portion:
forming n−1 products of the arrangement of weights with n−1 respective portions of the arrangement of features, the portion in a k-th product, of the n−1 products, being offset from the first portion by k+1 array elements in the first direction.

### Claim 16 (depends on 15)

The method of claim 15, wherein the processing circuit further comprises a cache, connected to the buffer and configured to supply features to the buffer, the cache having a size sufficient to store H+(H+n)*(W−1)−1 features, wherein:
H is a size of the arrangement of weights in the first direction, and
W is a size of the arrangement of weights in the second direction.

### Claim 17 (depends on 11)

The method of claim 11, wherein:
the buffer is configured to include:
a first queue connected to the first multiplier, and
a second queue connected to the second multiplier,
the first queue comprises a first register and a second register adjacent to the first register, the first register being an output register of the first queue,
the first multiplier is further configured:
in a first state:
to multiply a first weight by an activation from the output register of the first queue, and
in a second state:
to multiply the first weight by an activation from the second register of the first queue.

### Claim 18 (depends on 17)

The method of claim 17, wherein, in the second state, the output register of the first queue contains zero.

### Claim 19 (depends on 17)

The method of claim 17, wherein the processing circuit further comprises a first adder,
the method further comprising, in the first state:
connecting the first adder to:
an output of the first multiplier, and
an output of the second multiplier, and
adding, by the first adder:
a product received from the output of the first multiplier, and
a product received from the output of the second multiplier.

### Claim 20

A method for calculating with a means for processing, the means for processing comprising:
a register for storing an arrangement of weights;
a buffer for storing an arrangement of features;
a first multiplier; and
a second multiplier;
the method comprising performing a convolution of an arrangement of features with an arrangement of weights, the performing of the convolution comprising:
forming a product of the arrangement of weights with a first portion of the arrangement of features;
forming a product of the arrangement of weights with a second portion of the arrangement of features, the second portion being offset from the first portion by n arrangement of elements in a first direction, n being a positive integer; and
forming a product of the arrangement of weights with a third portion of the arrangement of features, the third portion being offset from the second portion by one arrangement of elements in a second direction, perpendicular to the first direction,
wherein the second portion and the third portion are spaced apart from an end of a row of the arrangement of features.
