Matrix Operation Metering Circuit for Charge-Based Current Summation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing vector-matrix multiplier circuits face inefficiencies in processing large matrices due to the need for separate analog-digital converters and potential rounding errors in current integration, especially in computationally intensive tasks like artificial intelligence applications.
Innovation Solution
An electricity metering circuit integrates output currents over time, using a series of stages to convert input currents into charge units, eliminating the need for separate analog-digital converters and allowing for the summation of output vectors from successive input vectors, with semiconductor switching elements enabling the processing of large matrices by breaking them into parts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate analog-digital converters are used for each column line in a vector-matrix multiplier, then accurate digital output can be obtained, but the device complexity and cost increase significantly
Solution Approach 1:
The patent combines multiple analog-digital conversion functions into a single shared converter. By using an integrator to accumulate charges from multiple column lines and a single analog-digital converter to convert the accumulated charge, the system achieves accurate digital output for multiple columns while requiring only one converter, thereby reducing device complexity and cost
Solution Approach 2:
The single analog-digital converter is designed to serve multiple column lines universally. The integrator accumulates charges from any combination of column lines, and the shared converter converts this accumulated charge to digital output, making the converter a multi-functional component that handles conversion for the entire matrix operation circuit
2Adaptability or versatility
If the matrix operation circuit processes different input voltage vectors sequentially, then the same hardware can handle multiple computations, but the processing speed decreases due to sequential operation
Solution Approach 1:
The integrator maintains continuous operation by accumulating charges from successive input vectors without interruption. Rather than resetting between computations, the integrator continuously integrates charges, allowing the system to process multiple input vectors in a continuous manner, thereby maintaining high adaptability while improving processing throughput
Solution Approach 2:
The integrator performs preliminary charge accumulation during the input vector application period. By pre-integrating the charges while the inputs are being applied, the system prepares the accumulated charge for immediate conversion to digital output, reducing idle time and improving overall processing speed
3Measurement precision
If the integrator stores large amounts of charge for multiple input vectors, then accurate summation is achieved, but the storage charge capacity requirements increase
Solution Approach 1:
The system employs periodic discharge of the integrator at regular intervals. By accumulating charges during one period and then discharging to a reference level, the integrator maintains operation within a limited charge range. This periodic action allows accurate summation over multiple input vectors while keeping the required storage charge capacity manageable
Solution Approach 2:
The system uses feedback from the analog-digital converter output to control the integrator discharge. When the accumulated charge reaches a level corresponding to a full-scale digital output, the system triggers a discharge to a reference level, ensuring the integrator operates within optimal charge boundaries while maintaining measurement precision through controlled accumulation cycles
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach reduces the need for additional components and minimizes rounding errors, enabling efficient processing of large matrices with accurate summation results, particularly suitable for artificial intelligence applications.
Implementation Method 1
an integrator stage set up to store the storage charge and to provide a second voltage proportional to the storage charge, the charge current being supplied to the integrator stage to increase the storage charge
Data Source
AI summary
An electricity metering circuit for a matrix operation circuit, having a circuit input for an electrical input current that is an output current of the matrix operation circuit. The electricity metering circuit is set up to provide a ground potential at the circuit input, to integrate the input current at the circuit input over time, to store a storage charge that is increased up to a predetermined maximum storage charge in accordance with a proportionality constant, proportionally to the integrated input current, to quantify the integrated input current in a charge unit, the charge unit corresponding to the maximum storage charge taking into account the proportionality constant, and to determine the integrated input current rounded down to the nearest integer charge unit as a count sum.