Neural Network MPC Decoding for Low-Power ECC Hardware
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing error correction coding (ECC) decoding techniques in memory devices and wireless baseband circuitry require significant computational resources, leading to increased semiconductor area and power consumption, which is undesirable in emerging applications like IoT and tactile internet where low power and low latency are crucial.
Innovation Solution
The use of multi-layer neural networks and recurrent neural networks to implement Message Probability Computing (MPC) decoders, which eliminate the need for multiple computational passes and feedback error paths, reducing semiconductor area and power consumption by estimating message probability compute data equivalent to iterative decoding processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional iterative decoding techniques are used, then decoding reliability is improved, but semiconductor area and power consumption increase
Solution Approach 1:
The patent transforms the iterative decoding process into a single-pass operation by changing the computational parameters. Instead of performing multiple iterative passes with feedback loops, the system uses a trained neural network model that performs decoding in one forward pass, fundamentally altering the computational parameters from iterative to direct evaluation.
Solution Approach 2:
The neural network model is trained offline in advance using iterative decoding algorithms. This preliminary training phase captures the complex iterative decoding behavior in pre-computed weights and biases, allowing the deployed system to perform decoding without actual iterative loops, thus reducing runtime computational requirements.
2Reliability
If traditional iterative decoding techniques are used, then decoding reliability is improved, but device complexity increases
Solution Approach 1:
The patent extracts the complex iterative decoding logic from the runtime system and relocates it to the training phase. The intricate feedback loops and multiple passes are removed from the operational decoder, leaving only a simple neural network forward pass implementation that achieves the same decoding reliability.
Solution Approach 2:
The patent creates a simplified copy of the iterative decoding functionality through the neural network model. Instead of implementing the actual complex iterative process, a trained network copy performs the same function through pattern recognition learned during training, reducing hardware complexity while maintaining performance.
3Measurement precision
If traditional iterative decoding techniques are used, then decoding accuracy is improved, but processing time increases
Solution Approach 1:
The neural network is trained in advance using high-accuracy iterative decoding algorithms. This preliminary action pre-computes the optimal decision boundaries and probability estimates, allowing the deployed system to achieve the same accuracy without performing time-consuming iterative passes during actual decoding operations.
Data Source
AI summary
Examples described herein utilize multi-layer neural networks, such as multi-layer recurrent neural networks to estimate message probability compute data based on encoded data (e.g., data encoded using one or more encoding techniques). The neural networks and/or recurrent neural networks may have nonlinear mapping and distributed processing capabilities which may be advantageous in many systems employing a neural network or recurrent neural network to estimate message probability compute data for a message probability compute (MPC) decoder. In this manner, neural networks or recurrent neural networks described herein may be used to implement aspects of error correction coding (ECC) decoders, e.g., an MPC decoder that iteratively decodes encoded data.