Changeflow GovPing Telecom & Technology Google LLC Patent for Neural Accelerator Compiler
Routine Notice Added Final

Google LLC Patent for Neural Accelerator Compiler

Favicon for changeflow.com ChangeBridge: Patent Grants - AI & Computing (G06N)
Published March 24th, 2026
Detected March 24th, 2026
Email

Summary

The USPTO has granted Google LLC a patent for a neural accelerator compiler designed to improve runtime speed and conserve energy. The patent, filed in March 2020, describes a method for optimizing the compilation and loading of neural network models into hardware accelerators.

What changed

The United States Patent and Trademark Office (USPTO) has issued patent US12585443B2 to Google LLC for a novel compiler system for neural accelerators. This patent, filed on March 9, 2020, and granted on March 24, 2026, details a compiler that identifies frequently invoked sequences of neural network models, compiles them, and loads them into a hardware accelerator's SRAM. The system aims to prevent unnecessary reloads by checking for existing compiled models, thereby enhancing runtime speed and reducing computational energy consumption.

This patent grant is primarily an intellectual property development for Google LLC and does not impose new regulatory obligations on other entities. However, it signifies advancements in AI hardware optimization that may influence future industry standards and competitive landscapes in the technology sector. Compliance officers in technology companies should be aware of such innovations as they may impact product development roadmaps and intellectual property strategies.

Source document (simplified)

← USPTO Patent Grants

Compiler for neural accelerator

Grant US12585443B2 Kind: B2 Mar 24, 2026

Assignee

Google LLC

Inventors

Arun Chauhan, Raksit Ashok, Dong Hyuk Woo

Abstract

A compiler of a computing device is described that identifies a sequence of neural network models frequently invoked by an application of the computing device, compiles the models in that sequence, and loads a static random access memory (SRAM) of a hardware accelerator with the compiled models only when the same compiled models—from another, but same, sequence that was previously invoked—are not already present in the SRAM. This prevents unnecessary reloading of compiled models into the SRAM, thereby increasing runtime speed and conserving computational energy.

CPC Classifications

G06F 8/41 G06N 3/06 G06N 3/063 G06N 3/045 G06N 3/105 G05B 2219/13119 G05B 2219/23266

Filing Date

2020-03-09

Application No.

17637190

Claims

20

View original document →

Named provisions

Compiler for neural accelerator

Classification

Agency
USPTO
Published
March 24th, 2026
Instrument
Notice
Legal weight
Non-binding
Stage
Final
Change scope
Minor
Document ID
US12585443B2

Who this affects

Applies to
Technology companies
Industry sector
5112 Software & Technology
Activity scope
AI Model Optimization
Geographic scope
United States US

Taxonomy

Primary area
Intellectual Property
Operational domain
IT Security
Topics
Artificial Intelligence Computer Science

Get Telecom & Technology alerts

Weekly digest. AI-summarized, no noise.

Free. Unsubscribe anytime.

Get alerts for this source

We'll email you when ChangeBridge: Patent Grants - AI & Computing (G06N) publishes new changes.

Free. Unsubscribe anytime.