Neural Network ADC Scaling with ReRAM Crossbar Arrays
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural network accelerators using resistive random-access memory (ReRAM) crossbar arrays face challenges in efficiently scaling signals during matrix-vector multiplication operations, leading to high overhead in area, energy, and power consumption due to large analog-to-digital converter (ADC) components.
Innovation Solution
The implementation of a neural network apparatus with a ReRAM crossbar array structure, featuring a first ADC scaler that adjusts the reference voltage in the analog domain and a second ADC scaler that multiplies the digital output signal by a shared scale factor in the digital domain, reducing ADC overhead while maintaining high accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If large ADC components are used to maintain high calculation accuracy, then measurement precision is improved, but area and power consumption increase
Solution Approach 1:
The patent segments the ADC functionality by introducing separate scaling stages (first ADC scaler and second ADC scaler) that divide the conversion process into multiple steps. This allows the main ADC to be smaller while achieving the same overall precision through coordinated scaling operations, directly resolving the contradiction between accuracy and area.
Solution Approach 2:
The patent changes the operating parameters of the ADC system by introducing scale factors (first scale factor and second scale factor) that dynamically adjust the reference voltage and digital output. This parameter transformation enables the ADC to achieve high measurement precision with a reduced physical size, as the scaling operations compensate for the smaller resolution of the individual ADC components.
2Measurement precision
If large ADC components are used to maintain high calculation accuracy, then measurement precision is improved, but power consumption increases
Solution Approach 1:
The patent segments the power consumption by dividing the ADC function into multiple low-power scaling stages rather than one high-power high-resolution ADC. The first ADC scaler and second ADC scaler operate with lower individual power requirements, and their combined effect achieves the same accuracy as a single large ADC, thereby reducing total power consumption.
Solution Approach 2:
The patent changes the power consumption characteristics by introducing scale factors that enable the system to use smaller, lower-power ADC components. The reference voltage scaling and digital output scaling operations allow the system to achieve high precision without requiring the high power consumption of large-resolution ADCs, thus resolving the contradiction between accuracy and power usage.
3Device complexity
If signal scaling is performed in the analog domain only, then device complexity is reduced, but measurement precision deteriorates
Solution Approach 1:
The patent adds a digital domain dimension to the scaling operations by introducing a second ADC scaler that performs digital scaling after the analog-to-digital conversion. This dual-domain approach (analog scaling + digital scaling) achieves higher measurement precision than analog-only scaling while keeping the overall device complexity manageable through the modular architecture, resolving the contradiction between complexity and precision.
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 significantly reduces the area and power consumption of ADC circuits while maintaining high calculation accuracy, optimizing the neural network hardware for efficient signal scaling and processing.
Implementation Method 1
the first ADC scaler may be configured to adjust a reference voltage corresponding to the reference signal by dividing the reference voltage by a scale factor in an analog domain
Implementation Method 2
the second ADC scaler may be configured to adjust the digital output signal by multiplying the digital output signal by the scale factor in a digital domain
Implementation Method 3
an analog-to-digital converter (ADC) circuit configured to generate a digital output signal based on a reference signal and the analog output signal of the RAM
Implementation Method 4
a random-access memory (RAM) configured to generate an analog output signal based on an input and a weight of a neural network, the RAM including a crossbar array structure
Data Source
AI summary
An apparatus includes: a random-access memory (RAM) configured to generate an analog output signal based on an input and a weight of a neural network, the RAM including a crossbar array structure; an analog-to-digital converter (ADC) circuit configured to generate a digital output signal based on a reference signal and the analog output signal of the RAM; a first ADC scaler configured to scale the reference signal of the ADC circuit; and a second ADC scaler configured to scale the digital output signal generated by the ADC circuit.