Charge-Trapped Transistor Neural Circuits for Low-Power Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural networks face challenges in power efficiency and cost due to reliance on power-hungry processors and unstable analog in-memory computing solutions, which are impractical for real-world applications with limited power budgets.
Innovation Solution
The implementation of an analog processing system using Charge-Trapped Transistors (CTTs) for neural network processing, where calculations are performed ratiometrically in time, eliminating dependencies on absolute voltage or current, and utilizing a Threshold Inverter Quantization (TIQ) comparator for stable and efficient signal processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If conventional processors (CPUs/GPUs) are used for neural network computation, then computational capability is achieved, but power consumption is excessively high
Solution Approach 1:
The patent replaces conventional digital processing systems with an analog neural network processor that directly implements neural network operations through analog circuitry. The system uses analog multipliers to perform multiplication operations and analog adders to perform accumulation operations, eliminating the need for digital-to-analog conversion and subsequent digital processing. This substitution of analog for digital mechanics in the computational path significantly reduces power consumption while maintaining the required computational capability for neural network inference.
2Use of energy by stationary object
If analog in-memory computing is implemented, then power consumption is reduced, but stability and reliability deteriorate
Solution Approach 1:
The patent incorporates feedback mechanisms through calibration circuits that continuously monitor and adjust the operation of analog multipliers and adders. The system includes reference voltage generators and trimming circuits that provide feedback to maintain optimal operating points, compensating for drift and variability in analog components. This feedback control ensures stable and reliable operation of the analog neural network processor while maintaining low power consumption.
3Measurement precision
If digital implementation with absolute voltage representation is used, then computational accuracy is maintained, but power consumption increases
Solution Approach 1:
The patent changes the fundamental parameter representation from absolute voltage levels to differential voltage signals and analog current modes. The system uses differential signaling throughout the analog computation path, which provides inherent noise rejection and improved accuracy. The analog multipliers and adders operate with controlled voltage and current parameters that are optimized for low power consumption while maintaining computational precision through careful biasing and signal conditioning.
4Device complexity
If analog computation with absolute voltage or current dependencies is implemented, then circuit simplicity is achieved, but sensitivity to environmental variations increases
Solution Approach 1:
The patent implements equipotentiality through the use of differential signaling and balanced circuit topologies. The analog neural network processor uses differential pairs and balanced differential amplifiers that maintain equal and opposite signal paths, making the system immune to common-mode noise and environmental variations. The differential architecture ensures that both signal paths experience identical environmental conditions, canceling out the effects of temperature drift, supply voltage variations, and other environmental factors.
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 enables low-power, stable, and efficient neural network processing, reducing power consumption and silicon area usage while maintaining accuracy over long periods, making it suitable for everyday usage in resource-constrained environments.
Implementation Method 1
a gate dielectric for interposing between the gate and a substrate to which the CTT is applied. The gate dielectric is used to store trapped charge and adjust the threshold voltage of the CTT
Data Source
AI summary
Provided are computer systems, methods, and devices for operating an artificial neural network. The system includes neurons. The neurons include a plurality of synapses including charge-trapped transistors for processing input signals, an accumulation block for receiving drain currents from the plurality of synapses, the drain currents produced as an output of multiplication from the plurality of synapses, the drain currents calculating an amount of voltage multiplied by time, a capacitor for accumulating charge from the drain currents to act as short-term memory for accumulated signals, a discharge pulse generator for generating an output signal by discharging the accumulated charge during a discharging cycle, and a comparator for comparing an input voltage with a reference voltage. The comparator produces a first output if the input voltage is above the reference voltage and produces a second output if the input voltage is below the reference voltage.