MTJ Neural Network Memory Write Current for Lower Power Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural network devices face high power consumption due to large numbers of operations and memory accesses, necessitating a new approach beyond clock gating and power gating.
Innovation Solution
A neural network device with a memory system using MTJ elements and a power supply unit that sets a write current or voltage to a value that maintains processing accuracy while reducing power consumption by allowing controlled write errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If write current is set to specified value to ensure processing accuracy, then processing accuracy is maintained, but power consumption increases
Solution Approach 1:
The patent changes the write current parameter from the specified value to a lower set value that is smaller than the specified value but still maintains processing accuracy. This parameter optimization allows the system to achieve the same reliability with reduced energy consumption, directly resolving the contradiction between processing accuracy and power consumption.
2Use of energy by moving object
If write current is reduced to save power, then power consumption decreases, but processing accuracy may deteriorate
Solution Approach 1:
The patent identifies an optimal write current value that is lower than the specified value but sufficient to maintain processing accuracy. By carefully selecting this parameter, the system achieves reduced power consumption without sacrificing reliability, effectively resolving the contradiction between energy efficiency and processing accuracy.
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
Reduces power consumption in neural network devices by setting write currents below specified values, enabling further power savings through clock and power gating.
Implementation Method 1
a write current or a write voltage having a set value at which a processing accuracy of the neural network processing by the neural network operation unit is equal to or greater than a set accuracy and that is set to be smaller than a specified value determined to switch bits for the plurality of memory cells
Data Source
AI summary
Provided are a neural network device and an operating condition determination method capable of reducing power consumption. A neural network device 10 includes a processor 12, a power supply unit 14, and an operating condition determination unit 16. The processor 12 includes a logic operation circuit 21 and a memory 22, and performs, by a convolutional neural network, recognition processing on data to be processed. An output control circuit 32 of the power supply unit 14 adjusts a write current to a set value at which a recognition accuracy satisfies a preset set accuracy and that is set to be smaller than a specified value determined to switch bits in memory cells 24.