Spike Event Decision-Making Device for Neuromorphic Chips
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neuromorphic computing devices face inherent delays in processing spike events, leading to unreliable inference results and delayed decision-making, which is a challenge in achieving ultra-low latency and accuracy in Internet of Things (IoT) edge computing applications.
Innovation Solution
A spike event decision-making device and method that utilizes counting modules to determine decision-making results based on the number of spike events fired by neurons in a spiking neural network, allowing for adaptive decision-making without fixed time windows, and incorporating sub-counters to improve reliability and accuracy by considering transition rates and occurrence ratios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a fixed time window is used for decision-making, then the decision-making process is simple and structured, but processing delays increase and reliability decreases
Solution Approach 1:
The patent replaces fixed time windows with dynamic event-counting windows. The decision-making window adapts its duration based on the actual rate of spike events, expanding or contracting automatically. This dynamic approach eliminates the need to pre-set arbitrary time windows, reducing processing delays while maintaining decision-making reliability through adaptive sampling of neural outputs.
2Ease of operation
If a fixed time window is used for decision-making, then the processing logic is straightforward, but inference result reliability decreases
Solution Approach 1:
The system incorporates feedback mechanisms where the counting module continuously monitors spike event rates and adjusts the decision-making window accordingly. When spike rates are high, the window contracts to capture relevant events; when rates are low, it expands to accumulate sufficient data. This feedback-driven adaptation ensures reliable inference results across varying neural activity conditions without complicating the core processing logic.
3Loss of time
If event counting is used to determine decision timing, then processing delay is reduced, but system complexity increases
Solution Approach 1:
The counting module operates autonomously, automatically tallying spike events from the spiking neural network and triggering decision-making when predefined thresholds are reached. This self-service mechanism eliminates the need for external timing control or complex coordination between modules, reducing processing delay while keeping the added complexity minimal and localized to the counting function itself.
4Speed
If the decision-making window is shortened to reduce delay, then responsiveness improves, but insufficient spike events lead to inaccurate decisions
Solution Approach 1:
The decision-making window dynamically adjusts its size based on the observed spike event rate. For high-speed events with frequent spikes, the window contracts to maintain responsiveness. For slower events with sparse spikes, the window expands to accumulate enough events for accurate decision-making. This dynamic adaptation resolves the trade-off between speed and accuracy by optimizing the window size for each specific event stream.
Data Source
AI summary
The present disclosure relates to a spike event decision-making device, method, chip, and electronic device to eliminate inherent delays of an output spike event of the neuromorphic chip when reading decisions. The spike event decision-making device includes a first counting module configured to count a number of input spike events of the spiking neural network, a second counting module configured to count some or all of the output spike events of the spiking neural network; and a decision-making module configured to generate a decision-making result according to numbers of spike events fired by neurons in an output layer of the spiking neural network when the number counted by the first counting module reaches a first predetermined value, or when the total count counted by the second counting module reaches a second predetermined value.