ANN Weight Safety Mechanism Using CRC Error Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Artificial neural networks (ANNs) are prone to random errors and directed deception, especially in critical applications like autonomous vehicles, necessitating mechanisms for functional safety to enhance system immunity and detect errors promptly.
Innovation Solution
The implementation of safety mechanisms in an ANN processor, including data stream fault detection, software-defined redundant allocation, cluster interlayer safety, layer control unit instruction addressing, weights safety, and neural network intermediate results safety, which are designed to detect and correct errors, ensuring system-level safety and compliance with standards like ISO 26262.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If safety mechanisms are implemented in ANN processor, then reliability is improved, but device complexity increases
Solution Approach 1:
The patent implements preliminary action by calculating and storing CRC checksums of weights before they are used in neural network operations. The weights are protected with checksums in advance, and the verification occurs during weight loading rather than during computation, preventing errors before they propagate through the system.
Solution Approach 2:
The patent introduces CRC checksums as an intermediary mechanism between weight storage and weight usage. This intermediary layer verifies weight integrity without requiring fundamental changes to the neural network computation architecture, thus improving reliability while minimizing complexity increase.
2Measurement precision
If CRC verification is performed on all weights, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The patent applies partial verification by performing CRC checks selectively on weights rather than continuously verifying all weights during computation. The verification is performed on the subset of weights that are loaded into the processor, balancing detection accuracy with time efficiency.
Solution Approach 2:
The CRC verification is performed preliminarily during weight loading phase rather than during the actual neural network computation phase. This timing strategy ensures that verified weights are used for computation, preventing re-verification during processing and minimizing time loss.
3Reliability
If redundant checksum storage is implemented, then reliability is improved, but loss of substance increases
Solution Approach 1:
The patent uses CRC checksums which are compact numerical representations (typically 16-32 bits) that provide comprehensive error detection capability. This parameter choice balances the need for reliable verification with the constraint of memory capacity, storing minimal redundant data while maintaining strong integrity checks.
Data Source
AI summary
Novel and useful system and methods of several functional safety mechanisms for use in an artificial neural network (ANN) processor. The mechanisms can be deployed individually or in combination to provide a desired level of safety in neural networks. Multiple strategies are applied involving redundancy by design, redundancy through spatial mapping as well as self-tuning procedures that modify static (weights) and monitor dynamic (activations) behavior. The mechanisms address ANN system level safety in situ, as a system level strategy tightly coupled with the processor architecture. The NN processor incorporates several functional safety concepts that function to detect and promptly flag and report an error with some mechanisms capable of correction as well. The safety mechanisms cover data stream fault detection, software defined redundant allocation, cluster interlayer safety, cluster intralayer safety, layer control unit (LCU) instruction addressing, weights storage safety, and neural network intermediate results storage safety.