Neural Network Power Distribution for Dynamic Load Switching

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing power management systems for dynamic loads, particularly neural network circuits, struggle to efficiently manage sudden and significant changes in power demand, leading to inefficiencies and potential thermal issues.

Innovation Solution

A power supply circuit comprising a voltage regulator and a distribution circuit that dynamically adjusts current output based on changes in the load, utilizing a distribution circuit with pass gates and feedback loops to optimize power delivery to segments of a neural network circuit.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If a voltage regulator is used to supply power to neural network circuits with dynamic load changes, then power delivery is maintained, but power management efficiency deteriorates due to inability to respond to sudden power demand changes

Engineering Contradiction:
Improvepower management efficiencyVSAvoidresponse to power demand changes
Core Design Contradiction:
Use of energy by moving objectVSAdaptability or versatility

Solution Approach 1:

The power supply circuit is divided into multiple parallel paths, each containing a switch element (first switch, second switch) that can be independently controlled. This segmentation allows selective activation of power paths based on load conditions, enabling efficient power management during inference operations by directing power only to active neural network segments.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The circuit employs dynamic switching mechanisms where the first and second switches are controlled based on real-time detection of power demand changes. The controller adjusts switch states in response to load conditions, transforming a static power supply into a dynamic system that adapts to varying power requirements during neural network inference.

Inventive Principle:
Principle #15Dynamics

2Reliability

If power is continuously supplied to neural network circuits, then operational readiness is maintained, but thermal stress increases due to unnecessary power delivery

Engineering Contradiction:
Improveoperational readinessVSAvoidthermal stress
Core Design Contradiction:
ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The circuit detects power demand changes in advance and preemptively adjusts switch states to prevent unnecessary power delivery. By monitoring load conditions and activating or deactivating power paths before excessive power consumption occurs, the system maintains operational readiness while preventing thermal stress from building up.

Inventive Principle:
Principle #9Preliminary anti-action

Solution Approach 2:

The system incorporates feedback mechanisms where the controller continuously monitors power demand and adjusts the state of the first and second switches accordingly. This closed-loop control ensures that power is supplied only when needed, maintaining operational readiness while minimizing thermal stress by eliminating unnecessary power delivery to inactive neural network segments.

Inventive Principle:
Principle #23Feedback

3Ease of operation

If a distribution circuit with multiple switches is used to manage power paths, then power delivery control is improved, but device complexity increases

Engineering Contradiction:
Improvepower delivery controlVSAvoidcircuit structure
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The circuit extracts and isolates the switching control function into a dedicated controller that manages the first and second switches independently. This separation of control logic from the power delivery paths simplifies the overall system architecture, making power delivery control more manageable despite the presence of multiple switches by centralizing intelligence in the controller.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentUS12353261B2Neural-network-based power management for neural network loads
Publication Date: 2025.07.08 QUALCOMM INC
  • US12353261B2 patent drawing
  • US12353261B2 patent drawing
  • US12353261B2 patent drawing

AI summary

Methods and apparatus for supplying power to a dynamic load, such as a neural network circuit. One example power supply circuit generally includes a voltage regulator circuit and a distribution circuit coupled to one or more outputs of the voltage regulator circuit. The distribution circuit is configured to output different amounts of current based on changes in the dynamic load. For certain aspects, the dynamic load includes a neural network circuit having a plurality of segments. In this case, the distribution circuit may be configured to output the different amounts of current based on which segment in the plurality of segments of the neural network circuit is active.