17833476. SUBTASK STORAGE FOR STREAMING CONVOLUTIONS IN NEURAL NETWORK PROCESSOR simplified abstract (Apple Inc.)

SUBTASK STORAGE FOR STREAMING CONVOLUTIONS IN NEURAL NETWORK PROCESSOR

Organization Name

Apple Inc.

Inventor(s)

Sayyed Karen Khatamifard of Bellevue WA (US)

Chenfan Sun of Shoreline WA (US)

Alon Yaakov of Raanana (IL)

Husam Khashiboun of Peqiin (IL)

Jeffrey D. Marker of Pleasant View UT (US)

Saman Naderiparizi of Seattle WA (US)

Ramana V. Rachakonda of Austin TX (US)

Rohit K. Gupta of Saratoga CA (US)

SUBTASK STORAGE FOR STREAMING CONVOLUTIONS IN NEURAL NETWORK PROCESSOR - A simplified explanation of the abstract

This abstract first appeared for US patent application 17833476 titled 'SUBTASK STORAGE FOR STREAMING CONVOLUTIONS IN NEURAL NETWORK PROCESSOR

Simplified Explanation

The patent application describes a method for streaming convolution operations in a neural processor circuit, which includes a neural engine circuit and a neural task manager.

• The neural task manager receives multiple task descriptors and subtask descriptors.
• Each task descriptor identifies a set of convolution operations for a specific layer in a set of layers.
• Each subtask descriptor identifies a task descriptor and a subset of convolution operations for a portion of a layer identified by the task descriptor.
• The neural processor circuit configures the neural engine circuit to execute the subset of convolution operations using the corresponding task descriptor.
• The neural engine circuit performs the subset of convolution operations and generates output data corresponding to input data of another subset of convolution operations identified by another subtask descriptor.

Potential applications of this technology:

• Neural processors for deep learning applications.
• Real-time image and video processing.
• Natural language processing tasks.
• Autonomous vehicles and robotics.

Problems solved by this technology:

• Efficient execution of convolution operations in neural networks.
• Streamlined processing of multiple layers and subsets of convolution operations.
• Improved performance and speed in neural network computations.

Benefits of this technology:

• Faster and more efficient execution of convolution operations.
• Enhanced performance and accuracy in neural network processing.
• Improved scalability for handling complex neural network models.
• Reduced power consumption and resource utilization in neural processors.

Original Abstract Submitted

Embodiments relate to streaming convolution operations in a neural processor circuit that includes a neural engine circuit and a neural task manager. The neural task manager obtains multiple task descriptors and multiple subtask descriptors. Each task descriptor identifies a respective set of the convolution operations of a respective layer of a set of layers. Each subtask descriptor identifies a corresponding task descriptor and a subset of the convolution operations on a portion of a layer of the set of layers identified by the corresponding task descriptor. The neural processor circuit configures the neural engine circuit for execution of the subset of the convolution operations using the corresponding task descriptor. The neural engine circuit performs the subset of the convolution operations to generate output data that correspond to input data of another subset of the convolution operations identified by another subtask descriptor from the list of subtask descriptors.