Optimal Bit Apportionments for Soft Error Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in handling multi-bit errors caused by nuclear radiation, particularly in noisy environments, where approaches like triple modular redundancy are insufficient and require excessive resources.
Innovation Solution
The method involves identifying optimal bit apportionments using simulations and a genetic algorithm to determine the number of bit copies for each data bit, allowing different bits to have varying numbers of copies, and using a voting scheme to estimate bit values when errors occur, potentially using both radiation-hardened and non-radiation-hardened memory.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If triple modular redundancy is used to handle soft errors, then reliability is improved, but device complexity and resource usage increase
Solution Approach 1:
The patent applies local quality by differentiating the treatment of individual bits within a data word. Instead of uniformly applying redundancy to all bits, the system identifies and applies redundancy only to specific bits that are more susceptible to errors or have higher impact on computational accuracy. This selective approach reduces overall redundancy requirements while maintaining reliability for critical bits.
Solution Approach 2:
The patent changes the parameter of redundancy from a fixed uniform value (3 copies for all bits in TMR) to a variable value that differs for each bit position. The redundancy parameter is optimized based on the specific computational function and error characteristics, allowing the system to adapt the number of copies to the actual needs of each bit rather than applying a blanket redundancy scheme.
2Reliability
If uniform redundancy is applied to all data bits, then reliability is improved, but resource usage increases
Solution Approach 1:
The patent implements local quality by assigning different redundancy levels to different bits based on their individual error susceptibility and importance. Critical bits that are more prone to errors or have greater impact on results receive higher redundancy, while less critical bits receive lower redundancy. This localized differentiation optimizes the distribution of memory resources.
Solution Approach 2:
The patent applies partial redundancy rather than excessive uniform redundancy. Instead of applying the same high level of redundancy to all bits, the system applies redundancy selectively and partially to only those bits that require it, based on error analysis and computational importance. This partial action approach reduces overall resource consumption while maintaining adequate error protection.
3Measurement precision
If more bit copies are stored for each data bit, then measurement precision is improved, but loss of substance increases
Solution Approach 1:
The patent applies local quality by concentrating memory resources on bits where precision is most critical rather than uniformly distributing redundancy across all bits. The system identifies which bits require higher accuracy based on the computational function and error characteristics, and applies additional copies only to those specific bits, thereby improving measurement precision where needed without proportionally increasing overall memory usage.
Data Source
AI summary
A method includes storing one or more bit copies of each of at least some bits of a data value in at least one memory. A number of bit copies of each bit of the data value is based on a specified apportionment, and different bits have different numbers of bit copies. The method also includes retrieving the bit copies of the at least some of the bits of the data value from the at least one memory. The method further includes, in response to determining that a specified bit of the data value has multiple retrieved bit copies that differ from one another, estimating a bit value for the specified bit using the multiple retrieved bit copies of the specified bit. In addition, the method includes outputting or using the data value having the estimated bit value for the specified bit.