Neuron Device With Time Division Multiplexed Synapse Modules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neuron devices lack efficient design flexibility and energy efficiency in processing and learning data, as they require multiple devices and complex configurations to achieve desired system functionality.
Innovation Solution
The neuron device incorporates an input unit, a synapse unit with multiple synapse modules operating in time division multiplexing mode, and an output unit, utilizing floating gate metal oxide silicon field effect transistors (MOSFETs) to apply coefficient information to input signals, enabling dynamic coefficient updates and weighted sum generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple neuron devices and synapses are used to configure a required system, then system functionality is achieved, but device complexity and configuration requirements increase
Solution Approach 1:
The neuron device is designed with multi-functional capabilities including input unit, synapse unit with multiple synapse modules, and output unit. Each synapse module contains multiple synapse elements that can be selectively activated, allowing a single device to perform multiple computational functions that would otherwise require multiple specialized devices.
Solution Approach 2:
The synapse unit is divided into multiple synapse modules, each containing multiple synapse elements. This segmentation allows selective activation of specific synapse elements based on computational requirements, reducing the need to configure and manage multiple complete neuron devices while achieving the same system functionality.
2Productivity
If conventional neuron devices are used for data processing and learning, then basic computational functions are achieved, but design flexibility and energy efficiency are insufficient
Solution Approach 1:
The neuron device employs dynamic coefficient updates through the synapse modules, where coefficient information can be modified based on learning requirements. The time division multiplexing mode allows dynamic switching between different synapse elements, enabling adaptive data processing that improves productivity while optimizing energy consumption by activating only necessary computational paths.
Solution Approach 2:
The time division multiplexing operation cycles through different synapse elements in sequential time slots. This periodic activation pattern allows the system to process data efficiently by distributing computational workload across multiple synapse elements over time, reducing peak energy consumption while maintaining high data processing capability.
3Adaptability or versatility
If multiple synapse elements are connected in series with time division multiplexing, then coefficient information can be dynamically applied to input signals, but operational complexity increases
Solution Approach 1:
The synapse modules are designed to automatically select and activate the appropriate synapse element based on the current time slot and computational requirements. This self-service mechanism eliminates the need for external control logic to manually manage the complexity of multiple synapse elements, allowing dynamic coefficient updates while maintaining operational simplicity.
Data Source
AI summary
A neuron device may include an input unit, a synapse unit, and an output unit. The synapse unit can be connected with the input unit and may include one or more synapse modules. Each of the one or more synapse modules may include multiple synapse elements connected in series and may be configured to operate in a time division multiplexing mode. Each synapse element may have specific coefficient information. In each of the one or more synapse modules, one of the multiple synapse elements connected in series may be configured to apply coefficient information to one of the multiple input signals received by the input unit. The output unit may obtain a weighted sum of the multiple input signals and may generate an output signal based on the weighted sum.