Tile-Based GPU Redundancy for Safety-Critical Graphics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing redundancy mechanisms in safety-critical systems, such as those in road vehicles, incur processing overhead due to the need for redundant hardware and software computations, consuming additional resources and power.
Innovation Solution
A tile-based graphics processing approach that enables redundancy per tile, allowing selective duplication and comparison of processing units, reducing the need for full hardware duplication and optimizing resource usage by enabling redundancy control through an Application Programming Interface (API).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If redundant hardware and software are used to achieve safety compliance, then reliability is improved, but processing overhead increases
Solution Approach 1:
The patent divides the graphics output into discrete tiles, allowing redundancy to be applied selectively to specific tile regions rather than the entire frame. This segmentation enables safety-critical areas to receive redundant processing while non-critical areas process normally, reducing overall computational overhead while maintaining required safety compliance.
Solution Approach 2:
Different regions of the graphics output are assigned different processing qualities based on their safety criticality. Tiles marked as redundant undergo duplicate processing and comparison, while non-redundant tiles process once. This local differentiation optimizes the balance between safety compliance and processing efficiency.
2Reliability
If full hardware duplication is implemented for redundancy, then reliability is improved, but device complexity increases
Solution Approach 1:
Instead of duplicating the entire graphics processing hardware pipeline, the patent implements redundancy through software-based tile duplication and comparison. The same processing operations are executed multiple times on selected tiles, and results are compared to detect mismatches. This approach achieves reliability through copying of computational results rather than physical hardware duplication.
Solution Approach 2:
The graphics processing hardware is designed to perform multiple functions: normal graphics processing and redundancy-based safety verification. The same execution units process both redundant and non-redundant tiles, and the hardware includes comparison capabilities that can operate on tile data without requiring separate dedicated redundancy hardware.
3Reliability
If redundant processing is applied to all data, then reliability is improved, but use of energy increases
Solution Approach 1:
Instead of applying redundant processing to all graphics data (excessive action), the patent implements partial redundancy where only selected tiles marked as safety-critical undergo duplicate processing. This partial application of the redundancy mechanism reduces overall energy consumption while maintaining safety compliance for the most critical portions of the output.
Data Source
AI summary
A method of operation of a tile-based graphics processor, including: receiving an instruction to process data to produce an output; detecting an indicator that a portion of the output is to be processed with redundancy; associating the indicator with at least one tile associated with the portion of the output; duplicating the processing of the data associated with the at least one tile by one or more execution units of the graphics processor to produce output data for each of a first and a second instance of the at least one tile; comparing output data for the first and the second instances of the at least one tile generated by the one or more execution units of the graphics processor; and responsive to detection of a mismatch between the output data of the first and the second instances, communicating a signal.