# US 2022/0044123 A1 · Embedded Programmable Logic Device for Acceleration in Deep Learning-Focused Processors

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

- Number: US 2022/0044123 A1
- Document ID: US20220044123A1
- Status: Published application
- Kind: A1
- Application number: 17/484,439
- Filed: 2021-09-24
- Published: 2022-02-10
- Earliest claimed: 2021-09-24
- Inventors: Rajesh Vivekanandham; Dheeraj Subbareddy; Dheemanth Nagaraj; Vijay S. R. Degalahal; Anshuman Thakur; Ankireddy Nalamalpu; MD Altaf Hossain; Mahesh Kumashikar; Atul Maheshwari
- Web page: https://patentgrep.com/patent/US20220044123A1

## CPC

- G06F 15/7867: {with reconfigurable architecture}
- G06F 9/30036: {Instructions to perform operations on packed data, e.g. vector, tile or matrix operations}
- G06F 15/8046: {Systolic arrays}
- G06F 2015/763: {ASIC}
- G06F 2015/768: {Gate array}
- G06N 3/063: using electronic means
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Processors may be enhanced by embedding programmable logic devices, such as field-programmable gate arrays. For instance, an application-specific integrated circuit device may include main fixed function circuitry operable to perform a main fixed function of the application-specific integrated circuit device. The application-specific integrated circuit also includes a support processor that performs operations outside of the main fixed function of the application-specific integrated circuit device, wherein the support processor comprises an embedded programmable fabric to provide programmable flexibility to application-specific integrated circuit device.

## Claims

### Claim 1

An application-specific integrated circuit device comprising:
main fixed function circuitry operable to perform a main fixed function of the application-specific integrated circuit device; and
a support processor that performs operations outside of the main fixed function of the application-specific integrated circuit device, wherein the support processor comprises an embedded programmable fabric to provide programmable flexibility to application-specific integrated circuit device.

### Claim 2 (depends on 1)

The application-specific integrated circuit device of claim 1, wherein the main fixed function comprises matrix multiplication.

### Claim 3 (depends on 2)

The application-specific integrated circuit device of claim 2, wherein the main fixed function circuitry comprises general matrix multiply circuitry.

### Claim 4 (depends on 2)

The application-specific integrated circuit device of claim 2, wherein the main fixed function circuitry comprises general matrix vector multiply circuitry.

### Claim 5 (depends on 1)

The application-specific integrated circuit device of claim 1, comprising memory.

### Claim 6 (depends on 5)

The application-specific integrated circuit device of claim 5, comprising a memory controller that controls the memory.

### Claim 7 (depends on 6)

The application-specific integrated circuit device of claim 6, wherein the memory controller comprises a memory controller-embedded programmable fabric.

### Claim 8 (depends on 7)

The application-specific integrated circuit device of claim 7, wherein the memory controller-embedded programmable fabric is physically outside of the memory controller but has access to internal functions of the memory controller.

### Claim 9 (depends on 1)

The application-specific integrated circuit device of claim 1, wherein the support processor comprises a tensor core.

### Claim 10 (depends on 1)

The application-specific integrated circuit device of claim 1 comprising a memory controller.

### Claim 11 (depends on 10)

The application-specific integrated circuit device of claim 10, wherein the memory controller comprises an additional embedded programmable fabric.

### Claim 12 (depends on 11)

The application-specific integrated circuit device of claim 11, wherein the embedded programmable fabric is configured to manipulate memory between stages of a neural network used to perform deep learning operations.

### Claim 13 (depends on 11)

The application-specific integrated circuit device of claim 11, wherein the embedded programmable fabric is configured to at least partially perform memory zeroing operations.

### Claim 14 (depends on 11)

The application-specific integrated circuit device of claim 11, wherein the embedded programmable fabric is configured to at least partially perform memory setting or other arithmetic operations.

### Claim 15 (depends on 1)

The application-specific integrated circuit device of claim 1 comprising a plurality of support processors including the support processor.

### Claim 16 (depends on 15)

The application-specific integrated circuit device of claim 15, wherein at least one other support processor of the plurality of support processors comprises an embedded programmable fabric.

### Claim 17

A method comprising:
performing a main fixed function in main fixed function circuitry of an application-specific integrated circuit device;
performing operations outside of the main fixed function of the application-specific integrated circuit device in a support processor; and
performing a subset of the operations outside of the main fixed function using an embedded programmable fabric embedded in the support processor to provide programmable flexibility to the application-specific integrated circuit device.

### Claim 18 (depends on 17)

The method of claim 17, wherein the subset of the operations comprises memory manipulation between layers or in a layer of a neural network performing deep learning operations.

### Claim 19

A system comprising:
a memory controller comprising an embedded programmable fabric, wherein the embedded programmable fabric is configured to perform memory manipulation between layers of a neural network performing deep learning operations; and
a processor comprising a systolic array and a processor-embedded programmable fabric, wherein the processor-embedded programmable fabric is configured to enhance deep learning operations using the system.

### Claim 20 (depends on 19)

The system of claim 19, wherein the processor comprises a graphics processing unit.
