Approximation Model Error Detection in Safety-Critical Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting errors in safety-critical systems, such as those in automotive applications, require significant hardware duplication and increased costs due to the need for fully redundant computations to ensure error detection, which is not efficient in terms of hardware expenditure.
Innovation Solution
A method and device utilizing at least two processing units, where one computes a main data model and the other computes an approximation data model, with a comparator unit to detect errors based on deviations between the two, allowing for reduced hardware expenditure by leveraging less powerful processing units for approximation computations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If fully redundant computation is used to detect errors in safety-critical systems, then error detection capability is improved, but hardware expenditure increases significantly
Solution Approach 1:
The patent applies partial action by computing only an approximation data model on the second processing unit rather than a full redundant computation. This partial computation provides sufficient error detection capability for safety-critical systems while significantly reducing the hardware resources required compared to complete redundancy.
Solution Approach 2:
The patent implements asymmetry by using two processing units with different capabilities - a first processing unit for main data model computation and a second, less powerful processing unit for approximation data model computation. This asymmetric configuration reduces hardware expenditure while maintaining error detection functionality.
2Reliability
If two identical processing units are used for redundant computation, then error detection reliability is improved, but silicon surface area increases
Solution Approach 1:
The patent reduces silicon surface area by employing asymmetric processing units - the second processing unit that computes the approximation data model is deliberately designed to be less powerful and occupy less silicon area than the first processing unit, while still providing sufficient error detection capability.
Solution Approach 2:
The approximation data model computation uses a simpler, less resource-intensive processing unit that occupies less silicon real estate. This 'cheaper' processing approach maintains error detection functionality while reducing the overall silicon surface area required for the system.
3Device complexity
If approximation data model computation is used instead of full redundant computation, then hardware expenditure is reduced, but measurement precision of error detection may be affected
Solution Approach 1:
The patent changes the parameter of computation accuracy by using an approximation data model instead of a full redundant model. This parameter change reduces hardware expenditure while the comparator unit maintains error detection precision by identifying deviations between the main and approximation results that indicate errors.
Solution Approach 2:
The comparator unit acts as an intermediary that bridges the main data model computation and the approximation data model computation. It detects errors by comparing results and determining deviations, thereby maintaining measurement precision despite the use of approximation computation.
Data Source
AI summary
A device for computing data models, in particular comprising the possibility to detect errors occurring during the computation, has at least two processing units, at least one of the at least two processing units being designed to compute a main data model as a function of at least one state of a system, at least one other of the at least two processing units being designed to compute, as a function of this at least one state of the system, an approximation data model associated with the main data model, the main data model comprising at least one property of the system as a first data model, the approximation data model comprising at least the same property of the system approximately as a second data model, a comparator unit being designed to compare a first result of a first computation of the main data model with a second result of a second computation of the approximation data model associated with the main data model, in order to determine information about a deviation between the first result and the second result, the comparator unit being designed to detect an error as a function of the information about the deviation if the deviation exceeds a maximum admissible deviation.