Neuromorphic Model State Control via Synaptic Time-Multiplexing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Developing practical, low-power integrated circuit implementations of large-scale neural and neuromorphic models for real-time processing and control systems that are energy efficient and scalable, while effectively mimicking brain functionality for applications like brain-computer interfaces and disease treatment.
Innovation Solution
The implementation of synaptic time-multiplexed neuromorphic networks with hardware-based circuits, which divide fully connected neural networks into decoupled sub-networks that operate in different time slots, allowing for a larger number of virtual connections with fewer physical synapses, and the use of feedback controllers to manage model states and control brain activity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If fully connected neural networks are implemented with hardware-based circuits, then processing power and brain functionality mimicry are improved, but the number of physical synapses and device complexity increase significantly
Solution Approach 1:
The patent divides the fully connected neural network into multiple decoupled sub-networks, where each sub-network processes a subset of connections. This segmentation allows the system to achieve full connectivity functionality while using fewer physical synapses by distributing connections across multiple specialized sub-networks rather than implementing all connections in a single monolithic structure.
Solution Approach 2:
The patent introduces a temporal dimension by operating different sub-networks in different time slots. Instead of requiring all connections to exist simultaneously in space, the system uses time-multiplexed operation where subsets of connections are activated sequentially, effectively trading spatial complexity for temporal processing.
2Device complexity
If the number of physical synapses is reduced, then device complexity and power consumption are decreased, but the number of virtual connections and processing capability are limited
Solution Approach 1:
The patent employs periodic operation where different sub-networks are activated in alternating time slots. Each sub-network is designed to handle specific connection subsets, and by periodically switching between them, the system achieves the functionality of a fully connected network with many more virtual connections than physical synapses, as each physical synapse is reused across multiple time periods.
Solution Approach 2:
The patent creates virtual copies of connection pathways by time-multiplexing the same physical hardware resources. Instead of building physical copies of all possible connections, the system uses a single set of physical synapses to simulate multiple connection pathways by activating different subsets at different times, effectively copying connection functionality through temporal multiplexing.
3Speed
If real-time processing is achieved with neuromorphic models, then control application responsiveness is improved, but energy efficiency and power consumption are worsened
Solution Approach 1:
The patent segments the neural network processing into multiple specialized sub-networks that can operate in parallel or sequential fashion. This segmentation allows real-time processing by distributing computational load across multiple smaller units rather than requiring a single large-scale network, reducing the energy burden on any single processing unit while maintaining overall real-time performance.
Solution Approach 2:
The patent uses periodic activation of different sub-networks in time slots, allowing the system to process information in discrete batches rather than continuous operation. This periodic action enables real-time processing responses while reducing average power consumption compared to continuously active full-network operation, as hardware resources are activated only when needed for specific processing tasks.
Data Source
AI summary
Model-based neural control uses a model of a portion of a brain and provides feedback control to the model that is based on a received output from the model. A neuromorphic model-based control system includes a neuromorphic model that includes a neuromorphic network to model the brain portion. A synaptic time-multiplexed (STM) neural model-based control system includes an STM neural network to the model the brain portion. The control systems further include a feedback controller to receive an output of the neuromorphic model or STM neural network and to provide a feedback control input to control a model state of the neuromorphic model or the STM neural network.