VECTOR COMPUTATION UNIT IN NEURAL NETWORK PROCESSOR

Title:

VECTOR COMPUTATION UNIT IN NEURAL NETWORK PROCESSOR

Document Type and Number:

Japanese Patent JP2023169224

Kind Code:

A

Abstract:

To provide a vector computation unit configured to enable parallel processing and pipelining of neural network computation thereby achieving high-speed execution.SOLUTION: An architecture 600 for executing neural network computation for a neural network including a plurality of layers comprises: activation circuitry 602 configured to receive a vector of accumulated values and configured to apply a function to each accumulated value to generate a vector of activation values; and normalization circuitry coupled to the activation circuitry and configured to generate a respective normalized value for each activation value. The normalization circuitry includes normalized register columns 604-610 which are equal in number to the number of columns in the activation circuitry or a systolic array.SELECTED DRAWING: Figure 6

More Like This:

WO/2022/136920	IMPROVED DISTRIBUTED TRAINING OF GRAPH-EMBEDDING NEURAL NETWORKS
WO/2023/073649	TECHNIQUES FOR OPTIMIZING NEURAL NETWORKS FOR MEMOIZATION USING SHIFTED VALUE LOCALIZATION
WO/2019/096660	COMPETITIVE MACHINE LEARNING ACCURACY ON NEUROMORPHIC ARRAYS WITH CAPACITOR MEMORY DEVICES

Inventors:

GREGORY MICHAEL THORSON
CHRISTOPHER AARON CLARK
LUU DAN

Application Number:

JP2023144224A

Publication Date:

November 29, 2023

Filing Date:

September 06, 2023

Export Citation:

Click for automatic bibliography generation Help

Assignee:

GOOGLE LLC

International Classes:

G06N3/063; G06F7/50; G06F7/523; G06F9/38; G06F17/16; G06G7/60

Attorney, Agent or Firm:

Patent Attorney Fukami Patent Office

Previous Patent: Resist composition and resist pattern manufacturing method

Next Patent: Context-sensitive control with radar-based gesture recognition