CIM Data Mapping for Low-Power Accurate ADC Conversion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computing in memory (CIM) circuits face high power consumption and reduced sampling precision due to the random distribution and wide range of to-be-processed array data, which negatively impacts energy efficiency.
Innovation Solution
A data processing method involving first and second mappings to reduce the numerical range of array data, followed by multiplier and accumulation operations, analog-to-digital conversion, and reverse mapping to restore output values, thereby reducing computational power consumption and maintaining conversion accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional CIM circuit processes randomly distributed and widely ranging array data directly, then the circuit can handle the full data range, but power consumption increases and ADC sampling precision deteriorates
Solution Approach 1:
The patent applies preliminary action by performing data mapping before the MAC operation to transform the input data into a standardized range. The mapping module converts widely ranging array data into a fixed range (e.g., 0-255), which then feeds into the MAC circuit. This preprocessing step ensures that the array circuit always operates on data within an optimal range, preventing power consumption spikes and maintaining ADC sampling precision without requiring the circuit to handle arbitrary large ranges.
Solution Approach 2:
The patent changes the parameter range of the input data through mapping operations. By transforming the data from its original widely ranging distribution into a standardized range, the system optimizes the operating parameters of the MAC circuit and ADC. This parameter transformation allows the circuit to operate in its optimal performance window, reducing power consumption while maintaining measurement precision.
2Productivity
If the numerical range of input data is wide and randomly distributed, then the data represents more information, but conversion time increases and energy efficiency decreases
Solution Approach 1:
The mapping operation performs preliminary action by pre-processing the input data to establish a consistent numerical range before conversion. This preprocessing step organizes the randomly distributed data into a structured format that the conversion circuit can process efficiently, thereby improving conversion speed without losing the essential information content of the original data.
Solution Approach 2:
The mapping module acts as an intermediary between the input data and the MAC-ADC conversion system. It transforms the widely ranging and randomly distributed input data into a standardized range that the conversion circuit is optimized to handle. This intermediary transformation enables fast conversion while preserving the relative relationships and information content of the original data through proportional mapping.
Data Source
AI summary
A data processing method and apparatus, an electronic device, and a storage medium are provided. The method includes: obtaining to-be-processed array data; performing first mapping on the to-be-processed array data to obtain target array data, and determining a weight corresponding to each target data in the target array data during the first mapping; performing multiplier and accumulation operation on the target data according to a first or second rule to obtain a first and second operation values; performing analog-to-digital conversion on the first operation value and the second operation value to obtain a first quantized value and a second quantized value, respectively; performing second mapping on the first quantized value and the second quantized value based on the weights to obtain a third quantized value and a fourth quantized value; and summing the third quantized value and the fourth quantized value to obtain a target output value.