DNN Layer Descriptor Fencing for Low-Power Dependency Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep neural networks (DNNs) require high computational complexity, leading to high power consumption, making them unsuitable for long-duration operations in low-power devices such as smartphones and power-over-Ethernet security cameras, where reduced power consumption is critical.
Innovation Solution
A neural network module that optimizes the flow of descriptors through a DNN module by executing layer descriptors in a manner that reduces latency, enabling faster processing and earlier power-down, thereby saving power.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If general-purpose processors like GPUs are used to implement DNN operations, then computational performance is improved, but power consumption increases
Solution Approach 1:
The patent segments the DNN processing workflow into discrete layer descriptors that can be independently managed and executed. Each descriptor represents a specific layer operation, allowing the system to process only necessary computations and enable early power-down when processing completes, thereby reducing overall power consumption while maintaining computational performance.
Solution Approach 2:
The patent implements preliminary action by pre-compiling layer descriptors and organizing them in a descriptor list before execution. This allows the DNN module to efficiently retrieve and execute pre-prepared instructions, reducing processing latency and enabling faster completion of DNN operations, which directly reduces power consumption in battery-powered devices.
2Reliability
If DNN processing is performed over long durations to achieve high accuracy, then classification performance is improved, but battery life decreases
Solution Approach 1:
The patent implements self-service through automatic descriptor management and flow optimization mechanisms within the DNN module. The system autonomously manages the execution flow of layer descriptors, optimizes processing sequences, and automatically determines when processing can be terminated, eliminating the need for continuous long-duration operation while maintaining classification accuracy.
Solution Approach 2:
The patent changes the parameter of processing duration by optimizing the execution flow of layer descriptors to complete processing faster. By managing descriptor flow efficiently and reducing latency between operations, the system achieves the same classification accuracy in shorter time, directly extending battery life in portable devices.
3Use of energy by moving object
If processing latency is reduced to enable faster power-down, then power consumption is decreased, but processing complexity increases
Solution Approach 1:
The patent introduces layer descriptors as intermediary structures that mediate between the DNN computation requirements and the hardware execution. These descriptors serve as intermediate representations that encode layer operations, enabling efficient flow management and latency reduction without significantly increasing overall system complexity, as the descriptors are systematically organized in a descriptor list.
Data Source
AI summary
A deep neural network (DNN) processor is configured to execute layer descriptors in layer descriptor lists. The descriptors define instructions for performing a forward pass of a DNN by the DNN processor. The layer descriptors can also be utilized to manage the flow of descriptors through the DNN module. For example, layer descriptors can define dependencies upon other descriptors. Descriptors defining a dependency will not execute until the descriptors upon which they are dependent have completed. Layer descriptors can also define a “fence,” or barrier, function that can be used to prevent the processing of upstream layer descriptors until the processing of all downstream layer descriptors is complete. The fence bit guarantees that there are no other layer descriptors in the DNN processing pipeline before the layer descriptor that has the fence to be asserted is processed.