IP Library Granted Patent US 12,061,969
Granted Patent B2
US 12,061,969 · App. 17/984,722 · Granted Aug 13, 2024

System and method for energy-efficient implementation of neural networks

Inventors: William Martin Snelgrove (Toronto, CA); Darrick Wiebe (Toronto, CA)
Assignee: UNTETHER AI CORPORATION
G06N3/045G06F9/3887G06F13/4022G06N3/063Y02D10/00
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,061,969
App. No.
17/984,722
Granted
Aug 13, 2024
Kind
B2
Abstract

A system and method for enhancing C*RAM, improving its performance for known applications such as video processing but also making it well suited to low-power implementation of neural nets. The required computing engine is decomposed into banks of enhanced C*RAM each having a SIMD controller, thus allowing operations at several scales simultaneously. Several configurations of suitable controllers are discussed, along with communication structures and enhanced processing elements.

Claims (12)

1. A single instruction, multiple data (SIMD) device comprising:

a plurality of banks, each bank including:

a controller including instruction memory; and

a plurality of processing elements with memory;

wherein the controller is connected to the plurality of processing elements with memory, the controller being operable to control the plurality of processing elements with memory to perform parallel operations, the instruction memory of the controller configured to be loaded from the memory of the plurality of processing elements.

2. The device of claim 1 , further comprising a decompressor operable to stream variable-length data from the instruction memory and produce fixed-length representations of coefficients.

3. The device of claim 1 , further comprising a decompressor operable to stream variable-length data from the instruction memory and produce instruction sequences.

4. A device comprising:

a plurality of banks, each bank including:

memory and a plurality of processing elements for performing parallel operations using the memory;

a single instruction, multiple data (SIMD) controller device, the SIMD controller device including instruction memory configured to be loaded from the memory of the plurality of processing elements;

wherein at least one of the SIMD controller devices includes a decompressor operable to stream variable-length data from instruction memory and produce decompressed information that includes one or both of fixed-length representations of coefficients and instruction sequences, the SIMD controller devices interconnected by a column bus to share the decompressed information.

Assignments (5)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Apr 23, 2026
From: UNTETHER AI CORPORATION
To: AT-MEMORY COMPUTING LP
Reel/Frame 075495/0905 →
RELEASE OF SECURITY INTEREST Recorded Jun 17, 2025
From: NATIONAL BANK OF CANADA
To: UNTETHER AI CORPORATION
Reel/Frame 071655/0897 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Nov 10, 2022
From: KAPIK INC.
To: UNTETHER AI CORPORATION
Reel/Frame 061726/0369 →
CONFIRMATION AND ASSIGNMENT Recorded Nov 10, 2022
From: SNELGROVE, WILLIAM MARTIN
To: KAPIK INC.
Reel/Frame 061917/0700 →
CONFIRMATION AND ASSIGNMENT Recorded Nov 10, 2022
From: WIEBE, DARRICK
To: KAPIK INC.
Reel/Frame 061917/0722 →
Continuity (3)
Division 15903754 · Feb 23, 2018
Provisional Application 62463318 · Feb 24, 2017
Related Publication 20230069360A1 · Mar 2, 2023