Silent Data Corruption by 10x Test Escapes Threatens Reliable Computing
12 days ago
- #Computer Science
- #Silent Data Corruption
- #Hardware Architecture
- Too many defective compute chips are escaping manufacturing tests, exceeding industrial targets by at least 10x.
- Silent data corruptions (SDCs) from test escapes threaten reliable computing.
- A three-pronged approach is proposed: quick diagnosis of defective chips from system-level behaviors, in-field detection, and new test experiments.
- New test experiments must address drawbacks of previous industrial tests and case studies.