Analog In-Memory MAC Conversion for Accurate Neural Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Digital computing for deep neural networks faces challenges with high energy consumption, large size, and high cost, limiting its application in various fields due to limitations in parallel processing and memory barriers, while analog in-memory computing offers a solution for large-scale parallel processing with reduced energy usage but requires accurate digitization of analog current signals.
Innovation Solution
An analog computing apparatus and method that includes a memory array with non-volatile memory cells, a digital-analog converter, a multiplexer, and an analog-digital converter, which classifies current signals by grade and converts them into digital signals using optimized I-V converters and ADCs with varying resistances or capacitances to maintain high accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If digital operator is used for deep neural network, then high accuracy is achieved, but energy consumption increases and device size becomes large
Solution Approach 1:
The patent replaces digital computing systems with analog computing systems to perform MAC operations. The analog computing apparatus uses continuous physical quantities (voltages, currents) to represent and process data, substituting the discrete digital logic operations with analog signal processing. This substitution enables parallel processing of multiple neurons simultaneously, significantly reducing energy consumption while maintaining computational accuracy for neural network operations.
Solution Approach 2:
The patent changes the fundamental parameter representation from discrete digital values to continuous analog parameters (voltages and currents). By using analog signals with varying amplitudes to represent numerical values, the system can perform computations directly in the analog domain, avoiding the energy-intensive digital-to-analog conversion processes and enabling more efficient parallel processing architecture.
2Use of energy by moving object
If analog in-memory computing is used, then energy consumption is reduced and parallel processing is improved, but digitization accuracy of analog current signals deteriorates
Solution Approach 1:
The patent segments the analog computing system into distinct functional modules: memory array for weight storage, input unit with DAC for signal generation, output unit with multiplexer for signal selection, and control unit for coordination. This segmentation allows each module to be optimized for its specific function, particularly the output unit which is designed to maintain signal integrity during the critical analog-to-digital conversion process, thereby preserving digitization accuracy.
Solution Approach 2:
The patent introduces an intermediary control unit that manages the conversion process between analog and digital domains. The control unit coordinates the DAC in the input unit and the ADC in the output unit, ensuring that analog signals are properly conditioned and converted with minimal loss of information. This intermediary control mechanism bridges the analog computing core and digital processing systems, maintaining accuracy throughout the conversion process.
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
The apparatus significantly reduces errors in digitizing analog MAC operation results, expanding the applicability of analog computing by enhancing accuracy and reducing energy consumption.
Implementation Method 1
a memory array including a non-volatile memory cell in which a weight is stored
Implementation Method 2
convert the current signal into a digital signal through the analog-digital converter
Implementation Method 3
select a current signal output from each of a plurality of output lines of the memory array, which are the results of the MAC operation, one by one through the multiplexer
Data Source
AI summary
An analog computing apparatus for performing a multiply-accumulate (MAC) operation includes a memory array including a non-volatile memory cell in which a weight is stored, an input unit including a digital-analog converter (DAC), an output unit including a multiplexer and an analog-to-digital converter (ADC), and a control unit for controlling the memory array, the input unit, and the output unit, wherein the control unit may apply the input signal to the memory array such that the MAC operation is performed through the weight, select a current signal output from each of a plurality of output lines of the memory array, which are the results of the MAC operation, one by one through the multiplexer, classify the grade of the selected current signal according to a current size, control the current signal according to the classified grade, and convert the current signal into a digital signal through the analog-digital converter.