Spiking Neural Network Data Processing Device for Low Power Time-Series Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data processing devices for spiking neural networks struggle to efficiently extract features from time-series data and convert them into spike signals, which is crucial for quick data processing and low power operation.
Innovation Solution
A data processing device that includes a discretizer to sample time-series data and a control unit to extract voltage and time features, converting these features into spike signals by increasing the number of spikes in corresponding input neurons based on threshold voltages and normalized time values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If voltage features and time features are extracted from sampled time-series data to increase the number of spikes firing in input neurons, then the capability to process time-series data quickly and with low power consumption is improved, but the device complexity increases due to the need for discretizer, control unit, voltage comparator, counter, and normalization device
Solution Approach 1:
The data processing device is divided into distinct functional modules: a discretizer that samples time-series data, a control unit that extracts voltage and time features, a voltage comparator that detects threshold crossings, a counter that measures time intervals, and a normalization device that scales time values. This segmentation allows each component to perform a specific function efficiently, enabling quick data processing while maintaining modularity that manages complexity.
Solution Approach 2:
The patent transforms time-series data into a different dimensional representation by converting continuous voltage and time features into discrete spike signals. The voltage comparator detects when voltage passes between threshold voltages, and the counter measures time values, which are then normalized to create spike patterns in the temporal domain. This dimensionality change from continuous analog signals to discrete spike events enables low-power processing in the spiking neural network domain.
2Measurement precision
If the number of spikes firing in input neurons is increased based on voltage feature and time feature extraction, then the feature extraction capability is improved, but the energy consumption increases due to additional processing operations
Solution Approach 1:
The patent replaces traditional continuous analog signal processing with discrete event-based spike processing. Instead of continuously processing analog voltage signals, the system uses a voltage comparator to detect discrete threshold crossing events and a counter to measure discrete time intervals between events. This substitution of continuous mechanical-like processing with discrete event processing reduces energy consumption while maintaining precise feature extraction capability through the spike timing and frequency patterns.
Solution Approach 2:
The normalization device changes the parameter scale of time values by multiplying them with a normalization factor to convert them into normalized time values within a specific range. This parameter transformation allows the system to work with standardized spike timing values that are energy-efficient to process while preserving the precise temporal relationships in the original data, thus maintaining measurement precision without proportionally increasing energy consumption.
3Measurement precision
If voltage comparator detects voltage passing between multiple threshold voltages and spike number allocator increases spikes in corresponding neurons, then the measurement precision of voltage features is improved, but the device complexity increases
Solution Approach 1:
The voltage range is segmented into multiple discrete threshold levels (first to m-th threshold voltages), and the spike number allocator is segmented to handle different threshold crossing events. The voltage comparator is divided into comparison logic that checks against each threshold independently. This segmentation enables precise voltage feature measurement by detecting which specific threshold level is crossed, while the modular structure manages the complexity of handling multiple thresholds through systematic division of labor among components.
Data Source
AI summary
Disclosed is a data processing device of the spiking neural network, which includes a discretizer that receives time-series data, discretizes the time-series data based on sampling, and outputs sampled time-series data, and a control unit that receives the sampled time-series data, extracts a voltage feature and a time feature from the sampled time-series data, and increases the number of spikes firing in an input neuron each corresponding to the voltage feature and the time feature extracted from a plurality of input neurons including first to n-th input neurons, with respect to the “n”, which is an arbitrary positive integer.