Compensation Signal Generator for Neural Network Negative Input Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional artificial neural network operations requiring negative input signals increase circuit complexity and cost due to the need for subtracting two values, making it difficult to implement efficiently with high hardware requirements.
Innovation Solution
A data sensing device comprising a compensation signal generator, a weighting operator, and an arithmetic operator that generates a compensation signal to offset input signals, allowing for efficient implementation of artificial neural network operations with negative inputs using a simple hardware architecture.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional digital circuits or high-order processors are used to perform artificial neural network operations, then operation flexibility is improved, but hardware cost and power consumption increase
Solution Approach 1:
The patent replaces conventional digital circuits and high-order processors with a non-volatile memory-based computing system. The memory cells directly perform neural network operations through their electrical characteristics, substituting traditional computational hardware with a memory-based approach that reduces hardware complexity while maintaining operational flexibility.
2Device complexity
If non-volatile memory is used to perform artificial neural network operations, then hardware cost and power consumption are reduced, but the ability to handle negative input signals deteriorates due to increased circuit complexity
Solution Approach 1:
The patent applies preliminary action by pre-storing compensation data in the non-volatile memory that corresponds to negative input signals. Before actual neural network operations occur, the system prepares compensation values that can be directly applied during inference, eliminating the need for complex real-time subtraction circuits and enabling straightforward handling of negative inputs.
Solution Approach 2:
The patent introduces compensation data as an intermediary element between the input signal and the neural network computation. This compensation data acts as a mediator that transforms negative input handling into a simple addition operation, avoiding the need for complex subtraction circuitry while maintaining computational accuracy.
3Adaptability or versatility
If subtraction operations are implemented for negative input signals by using two input values, then negative input capability is achieved, but the number of operations triples and circuit cost increases
Solution Approach 1:
The patent pre-calculates and stores compensation values in the non-volatile memory that correspond to negative input scenarios. During actual operation, the system simply adds these pre-computed compensation values to the input, transforming a complex three-step subtraction process into a single addition operation and thereby tripling operational efficiency.
Solution Approach 2:
The patent changes the computational parameter from subtraction (which requires multiple operations) to addition (which requires a single operation). By transforming the mathematical operation type and pre-computing the necessary compensation values, the system maintains negative input capability while reducing the number of operations from three to one.
Data Source
AI summary
A data sensing device and a data sensing method thereof are provided. The data sensing device includes a compensation signal generator, a weighting operator and an arithmetic operator. The compensation signal generator receives a basic input signal and a plurality of reference weighting values, and generates a compensation signal according to the basic input signal and the reference weighting values. The weighting operator has a plurality of memory cells, performs a writing operation on the memory cells according to the weighting values based on address information, and the weighting operator generates an output signal by the memory cells by receiving a plurality of input signals. The arithmetic operator performs an operation on the output signal and the compensation signal to generate a compensated output signal.