Frequency Match Circuit for Low-Power Pattern Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional devices for pattern recognition, such as artificial neural networks, are computationally intensive and slow due to their reliance on floating-point values and calculations, which require substantial circuitry demands.
Innovation Solution
The implementation of a Frequency Match Circuit (FMC) that utilizes combinatorial and sequential logic, operating within an arbitrary time window controlled by a master counter, to quickly identify patterns by generating oscillating signals based on input values and weight values, allowing for faster pattern recognition and reduced hardware requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional floating-point calculations are used for pattern recognition, then measurement precision is improved, but device complexity increases and processing speed decreases
Solution Approach 1:
The patent replaces traditional floating-point mechanical/electronic calculation systems with a neural network-inspired system that uses spike-timing-dependent plasticity (STDP) mechanisms. Instead of using complex floating-point arithmetic circuits, the system employs simplified neuron models where synaptic weights are adjusted based on the timing differences between pre-synaptic and post-synaptic spikes, fundamentally substituting the calculation paradigm to reduce circuitry complexity while maintaining pattern recognition capability
Solution Approach 2:
The patent changes the fundamental parameters of computation from continuous floating-point values to discrete spike timing events. By representing information in the temporal domain (timing of spikes) rather than magnitude domain (floating-point values), the system achieves pattern recognition with simpler circuitry that only needs to detect and respond to timing relationships, avoiding complex arithmetic operations
2Measurement precision
If traditional floating-point calculations are used for pattern recognition, then measurement precision is improved, but processing speed decreases
Solution Approach 1:
The patent implements periodic spiking activity where neurons fire at specific intervals based on their activation levels and synaptic weights. This periodic action allows the neural network to process information in discrete time steps rather than requiring continuous floating-point calculations, enabling faster parallel processing while maintaining the precision needed for pattern recognition through the temporal coding of information
Solution Approach 2:
The patent replaces sequential floating-point calculation mechanisms with parallel spike propagation mechanisms. Information is transmitted through the network via simultaneous spike events across multiple neurons, allowing parallel processing of pattern recognition tasks. The STDP learning rule updates synaptic weights asynchronously based on spike timing, eliminating the need for sequential arithmetic operations and dramatically increasing processing speed
3Measurement precision
If more hardware resources are allocated to traditional pattern recognition devices, then measurement precision is improved, but energy consumption increases
Solution Approach 1:
The patent implements self-organizing neural networks where synaptic weights automatically adjust through STDP mechanisms based on the temporal correlation of spike events. The system learns and adapts to patterns in the input data without requiring external training computations or additional control hardware, achieving high precision pattern recognition through autonomous self-organization that minimizes energy consumption by eliminating dedicated training circuitry
Solution Approach 2:
The patent substitutes energy-intensive floating-point arithmetic operations with low-power spike detection and timing measurement circuits. The neural network performs computations using simple threshold comparisons and temporal delay measurements rather than complex mathematical operations, dramatically reducing power consumption while maintaining the ability to achieve high-precision pattern recognition through the collective behavior of spiking neurons
Data Source
AI summary
Digital data processing circuitry is described that uses combinatorial logic hardware and sequential logic hardware to process one or more inputs. For each input a periodic sequence is generated that is based at least in part on a value of the input and a weight value. A match is determined at an event time when the periodic sequence(s) matches a corresponding arbitrary reference pattern. The digital data processing circuitry may be implemented as an integrated circuit as part of an interconnected network of devices that may be trained and subsequently used for recognition or other complex data processing tasks.