# US 2026/0064799 A1 · Combinatorial optimization on tensor processors

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

- Number: US 2026/0064799 A1
- Document ID: US20260064799A1
- Status: Published application
- Kind: A1
- Application number: 19/313,166
- Filed: 2025-08-28
- Published: 2026-03-05
- Earliest claimed: 2024-08-29
- Assignees: Groq, Inc.
- Applicants: Groq, Inc.
- Inventors: Johannes Prosser; Zarko Bodroski; Gregory Reynolds Morse; Michael Booth; Peter Rakyta
- Web page: https://patentgrep.com/patent/US20260064799A1

## CPC

- G06F 17/11: for solving equations {, e.g. nonlinear equations, general mathematical optimization problems (optimization specially adapted for a specific administrative, business or logistic context G06Q10/04)}
- G06F 17/16: Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)}
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

The present disclosure relates to systems and methods for obtaining a problem specification file descriptive of a combinatorial optimization problem; identifying a plurality of combinable vector-matrix operations of the combinatorial optimization problem; generating an instruction set for the one or more processors, wherein the instruction set includes a first instruction that, when implemented, causes the one or more processors to perform a combined matrix-matrix operation replacing the plurality of combinable vector-matrix operations; and executing the instruction set.

## Claims

### Claim 1

A method, comprising:
obtaining, by a computing system comprising one or more processors, a problem specification file descriptive of a combinatorial optimization problem;
identifying, by the computing system, a plurality of combinable vector-matrix operations of the combinatorial optimization problem;
generating, by the computing system, an instruction set for the one or more processors, wherein the instruction set comprises a first instruction that, when implemented, causes the one or more processors to perform a combined matrix-matrix operation replacing the plurality of combinable vector-matrix operations; and
executing, by the computing system, the instruction set.

### Claim 2 (depends on 1)

The method of claim 1, wherein executing, by the computing system, the instruction set causes one or more matrix multiplication functional units of the one or more processors to perform the combined matrix-matrix operation.

### Claim 3 (depends on 1)

The method of claim 1, wherein the instruction set comprises a second instruction that, when implemented, causes the one or more processors to stream an output of the combined matrix-matrix operation to a vector multiplication functional unit of the one or more processors.

### Claim 4 (depends on 3)

The method of claim 3, wherein the output of the combined matrix-matrix operation comprises a partial vector result.

### Claim 5 (depends on 1)

The method of claim 1, further comprising converting the problem specification file from a first representation to a Quadratic Unconstrained Binary Optimization (QUBO) representation.

### Claim 6 (depends on 5)

The method of claim 5, wherein converting the problem specification file from the first representation to the QUBO representation comprises introducing one or more slack variables to convert an inequality constraint into an equality constraint.

### Claim 7 (depends on 5)

The method of claim 5, further comprising transforming the QUBO representation to an Ising spin Hamiltonian function.

### Claim 8 (depends on 5)

The method of claim 5, further comprising mapping the Ising spin Hamiltonian function to an optimization problem over continuous variables by replacing one or more spins in the Ising spin Hamiltonian function with harmonic oscillator functions.

### Claim 9 (depends on 1)

The method of claim 1, wherein generating, by the computing system, the instruction set for the one or more processors comprises batching the plurality of combinable vector-matrix operations.

### Claim 10 (depends on 1)

The method of claim 1, wherein the combinatorial optimization problem comprises a financial portfolio optimization problem.

### Claim 11

A system, comprising:
one or more processors; and
one or more non-transitory, computer-readable media storing instructions that, when implemented, cause the one or more processors to perform operations, the operations comprising:
obtaining a problem specification file descriptive of a combinatorial optimization problem;
identifying a plurality of combinable vector-matrix operations of the combinatorial optimization problem;
generating an instruction set for the one or more processors, wherein the instruction set comprises a first instruction that, when implemented, causes the one or more processors to perform a combined matrix-matrix operation replacing the plurality of combinable vector-matrix operations; and
executing the instruction set.

### Claim 12 (depends on 11)

The system of claim 11, wherein executing the instruction set causes one or more matrix multiplication functional units of the one or more processors to perform the combined matrix-matrix operation.

### Claim 13 (depends on 11)

The system of claim 11, wherein the instruction set comprises a second instruction that, when implemented, causes the one or more processors to stream an output of the combined matrix-matrix operation to a vector multiplication functional unit of the one or more processors.

### Claim 14 (depends on 13)

The system of claim 13, wherein the output of the combined matrix-matrix operation comprises a partial vector result.

### Claim 15 (depends on 11)

The system of claim 11, wherein the operations further comprise converting the problem specification file from a first representation to a QUBO representation.

### Claim 16 (depends on 15)

The system of claim 15, wherein converting the problem specification file from the first representation to the QUBO representation comprises introducing one or more slack variables to convert an inequality constraint into an equality constraint.

### Claim 17 (depends on 15)

The system of claim 15, wherein the operations further comprise transforming the QUBO representation to an Ising spin Hamiltonian function.

### Claim 18 (depends on 15)

The system of claim 15, wherein the operations further comprise mapping the Ising spin Hamiltonian function to an optimization problem over continuous variables by replacing one or more spins in the Ising spin Hamiltonian function with harmonic oscillator functions.

### Claim 19 (depends on 11)

The system of claim 11, wherein generating the instruction set for the one or more processors comprises batching the plurality of combinable vector-matrix operations.

### Claim 20

One or more non-transitory, computer-readable media storing instructions that, when implemented, cause one or more processors to perform operations, the operations comprising:
obtaining a problem specification file descriptive of a combinatorial optimization problem;
identifying a plurality of combinable vector-matrix operations of the combinatorial optimization problem;
generating an instruction set for one or more processors, wherein the instruction set comprises a first instruction that, when implemented, causes the one or more processors to perform a combined matrix-matrix operation replacing the plurality of combinable vector-matrix operations; and
executing the instruction set.
