Neural Network Circuit With Adaptive Precision Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network circuits face issues with meeting computation precision requirements in low precision scenarios and wasting power in high precision scenarios due to the fixed precision of DAC and ADC components.
Innovation Solution
A neural network circuit that adjusts output precision based on computation precision changes by using a sample-and-hold circuit, reference voltage generation, comparator circuit, and output circuit, with control signals to manage sampling frequency and duration, reducing component count and power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fixed precision DAC and ADC components are used in existing neural network circuits, then device complexity is reduced, but measurement precision cannot adapt to computation precision requirements
Solution Approach 1:
The patent applies dynamics by making the precision of DAC and ADC components adjustable rather than fixed. The system dynamically configures the precision of these components based on the computation precision requirements of different neural network layers, allowing the measurement precision to adapt to varying computational needs without requiring completely different hardware configurations for each precision level.
Solution Approach 2:
The patent implements parameter changes by allowing the precision parameters of DAC and ADC components to be modified according to the computation precision requirements. Different precision levels can be selected for different components based on the specific needs of the neural network computation, enabling the system to optimize both measurement precision and power consumption by matching component precision to actual computational requirements.
2Measurement precision
If high precision DAC and ADC components are used to meet computation precision requirements, then measurement precision is improved, but power consumption increases
Solution Approach 1:
The patent applies local quality by assigning different precision levels to different DAC and ADC components based on their specific functional requirements within the neural network circuit. Instead of using uniformly high precision across all components, the system configures each component's precision locally according to its role and the computation precision needs of the associated neural network layer, thereby reducing overall power consumption while maintaining necessary measurement precision.
Solution Approach 2:
The patent implements partial action by using high precision DAC and ADC components only where necessary to meet computation precision requirements, rather than applying high precision universally. Low precision components are used where sufficient accuracy can be achieved with lower power consumption, optimizing the balance between measurement precision and energy usage across the entire system.
3Use of energy by moving object
If low precision DAC and ADC components are used to reduce power consumption, then energy usage is reduced, but computation precision requirements cannot be met
Solution Approach 1:
The patent applies local quality by configuring each DAC and ADC component with the minimum necessary precision level required for its specific function and the associated neural network layer's computation precision requirements. This localized precision configuration ensures that power consumption is minimized while still meeting the computation precision needs of each specific component, avoiding both over-provisioning and under-provisioning of precision.
Solution Approach 2:
The patent implements partial action by using low precision components only where sufficient accuracy can be maintained at lower power consumption levels. The system carefully determines which components can operate at reduced precision without compromising overall computation precision requirements, thereby optimizing the trade-off between energy usage and measurement precision across the neural network circuit.
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 circuit adapts to varying precision needs, ensuring accurate computation results while minimizing power usage, thus overcoming precision mismatches and inefficiencies in existing systems.
Implementation Method 1
The first sample-and-hold circuit is connected to the first group of computation units, and is configured to generate a first analog voltage based on the first output current
Implementation Method 2
The first comparator circuit is connected to the first sample-and-hold circuit and the reference voltage generation circuit, and is configured to output a first level signal based on the first analog voltage and the reference voltage
Implementation Method 3
After being processed by a digital-to-analog converter (DAC), input data enters a computation array in a form of voltage. Subsequently, the voltage undergoes corresponding conductance to form currents and the currents converge on a same column.
Data Source
AI summary
A neural network circuit is described that includes a first sample-and-hold circuit, a reference voltage generation circuit, a first comparator circuit, and a first output circuit. The first sample-and-hold circuit generates a first analog voltage based on a first output current output by a first neural network computation array. The reference voltage generation circuit generates a reference voltage based on a first control signal. The first comparator circuit is connected to the first sample-and-hold circuit and the reference voltage generation circuit, and outputs a first level signal based on the first analog voltage and the reference voltage. The first output circuit samples the first level signal based on a second control signal, and outputs a first computation result that meets the first computation precision.