Direction-Selective Neuromorphic Circuits With Lateral Inhibition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neuromorphic architectures struggle to effectively address real-time detection of transient changes in space systems due to jitter and moving backgrounds, and event sensors face challenges in separating and reconstructing target temporal signatures amidst scene motion, with limitations in low SWaP and low false alarm rate.
Innovation Solution
Implementing direction-selective neuromorphic circuits using dendrites with inhibition or winner-takes-all mechanisms, leveraging CMOS transistors and non-volatile memory devices, to enhance spatiotemporal pattern recognition and direction selectivity, particularly for event sensors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If frame-based cameras are used for real-time detection, then detection coverage is comprehensive, but data volume increases 100× compared to event cameras
Solution Approach 1:
The patent extracts only the relevant temporal signature information from the full scene data by using event cameras that respond only to changes, rather than capturing complete frame data. This selective extraction reduces data volume while maintaining detection capability for transient events.
Solution Approach 2:
The patent segments the detection task into separate directional channels (e.g., upward, downward, leftward, rightward motion) using dedicated dendrite circuits for each direction. This segmentation allows efficient processing of different motion types independently, reducing overall computational burden.
2Quantity of substance
If event cameras are used to reduce data volume, then data efficiency improves, but ability to separate target temporal signature from scene motion deteriorates
Solution Approach 1:
The patent adds a directional dimension to the detection by implementing separate dendrite circuits for different motion directions. This dimensional expansion allows the system to distinguish target motion from background motion by analyzing which directional channel responds, solving the target separation problem.
Solution Approach 2:
Each dendrite circuit is tuned with specific weights to detect particular motion patterns in specific directions. This local specialization allows each circuit to excel at detecting its assigned direction while ignoring others, improving target separation through distributed specialized processing.
3Measurement precision
If directional selectivity is implemented through separate dendrite circuits, then pattern recognition accuracy improves, but circuit complexity increases
Solution Approach 1:
The patent merges multiple detection functions into a single neuromorphic circuit structure where dendrites with different weight configurations handle different directions. This consolidation achieves directional selectivity without requiring completely separate physical systems for each direction.
Solution Approach 2:
The patent uses dynamic weight adjustments in the dendrite circuits to adapt to different motion patterns. The weights can be programmed or learned to optimize detection for specific applications, allowing the same hardware structure to handle multiple detection scenarios.
4Reliability
If lateral inhibition is implemented for winner-takes-all detection, then false alarm rate decreases, but processing time increases due to inhibition propagation
Solution Approach 1:
The patent pre-configures the dendrite circuits with specific weight patterns that predispose them to respond to particular motion directions. This preliminary configuration reduces the computational burden during actual detection, allowing faster winner-takes-all decisions without extensive inhibition propagation.
Data Source
AI summary
A direction-selective neuromorphic circuit is provided comprising a first dendrite comprising first and second compartments and a destination compartment arranged sequentially, wherein the first dendrite is tuned to detect a first pattern. A second dendrite comprises first and second compartments and a destination compartment arranged sequentially, wherein the second dendrite is tuned to detect a second pattern. Input from a first spike generator is input to the first compartment of the first dendrite and the second compartment of the second dendrite. Input from a second spike generator is input to the first compartment of the second dendrite and the second compartment of the first dendrite. Responsive to detecting the first pattern, the destination compartment of the first dendrite spikes and laterally inhibits the second dendrite. Responsive to detecting the second pattern, the destination compartment of the second dendrite spikes and laterally inhibits the first dendrite.