Neuromorphic Circuit Customization for Low-Power ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional central processing units (CPUs) are unable to provide the necessary processing power for machine-learning applications while keeping power consumption low, particularly for custom processing capabilities in new applications and situations, especially in battery-powered devices.
Innovation Solution
A system and method for customizing neural networks on neuromorphic integrated circuits, which includes receiving user-specific target information, merging it with existing data, building a training set, and updating synaptic weights to determine custom processing capabilities for new applications and situations, allowing for efficient power usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional CPUs are used to process machine-learning applications, then processing power is improved, but power consumption increases
Solution Approach 1:
The patent replaces conventional digital CPU processing with a neuromorphic integrated circuit that mimics biological neural networks. This substitution enables the system to perform machine-learning operations with dramatically reduced power consumption (100x less than GPUs and 280x less than digital CMOS) while maintaining the necessary processing power for custom machine-learning applications.
Solution Approach 2:
The patent changes the fundamental operating parameters from clocked sequential processing to event-driven asynchronous processing. The neuromorphic circuit uses spike-based communication and continuous-time processing, fundamentally altering how computations are performed to achieve both high productivity and low energy consumption simultaneously.
2Adaptability or versatility
If conventional CPUs are used for custom processing capabilities, then adaptability is improved, but power consumption increases
Solution Approach 1:
The patent implements dynamic reconfigurability in the neuromorphic circuit, allowing the neural network architecture to be customized and updated for different machine-learning applications. The system can adapt its processing capabilities through programmable synaptic weights and configurable network topologies, providing versatility without the power penalty of conventional approaches.
3Use of energy by moving object
If neuromorphic integrated circuits are used, then energy efficiency is improved, but processing power decreases
Solution Approach 1:
The patent performs preliminary training of the neural network using conventional CPUs or GPUs to determine optimal synaptic weights. These pre-computed weights are then loaded into the neuromorphic circuit, which executes the actual machine-learning inference with high energy efficiency. This division of labor allows the system to achieve both high processing power during training and high energy efficiency during deployment.
Data Source
AI summary
Provided herein is a system including, in some embodiments, one or more servers and one or more database servers configured to receive user-specific target information from a client application for training a neural network on a neuromorphic integrated circuit. The one or more database servers are configured to merge the user-specific target information with existing target information to form merged target information in the one or more databases. The system further includes a training set builder and a trainer. The training set builder is configured to build a training set for training a software-based version of the neural network from the merged target information. The trainer is configured to train the software-based version of the neural network with the training set to determine a set of synaptic weights for the neural network on the neuromorphic integrated circuit.