|dc.description.abstract||Single Event Transient (SET) errors in ground-level electronic devices are a growing concern in the radiation hardening field. However, effective SET mitigation technologies which satisfy ground-level demands such as generic, flexible, efficient, and fast, are limited. The classic Triple Modular Redundancy (TMR) method is the most well-known and popular technique in space and nuclear environment. But it leads to more than 200% area and power overheads, which is too costly to implement in ground-level applications. Meanwhile, the coding technique is extensively utilized to inhibit upset errors in storage cells, but the irregularity of combinatorial logics limits its use in SET mitigation. Therefore, SET mitigation techniques suitable for ground-level applications need to be addressed.
Aware of the demands for SET mitigation techniques in ground-level applications, this thesis proposes two novel approaches based on the redundant wire and approximate logic techniques.
The Redundant Wire is a SET mitigation technique. By selectively adding redundant wire connections, the technique can prohibit targeted transient faults from propagating on the fly. This thesis proposes a set of signature-based evaluation equations to efficiently estimate the protecting effect provided by each redundant wire candidates. Based on the estimated results, a greedy algorithm is used to insert the best candidate repeatedly. Simulation results substantiate that the evaluation equations can achieve up to 98% accuracy on average. Regarding protecting effects, the technique can mask 18.4% of the faults with a 4.3% area, 4.4% power, and 5.4% delay overhead on average. Overall, the quality of protecting results obtained are 2.8 times better than the previous work. Additionally, the impact of synthesis constraints and signature length are discussed.
Approximate Logic is a partial TMR technique offering a trade-off between fault coverage and area overheads. The approximate logic consists of an under-approximate logic and an over-approximate logic. The under-approximate logic is a subset of the original min-terms and the over-approximate logic is a subset of the original max-terms. This thesis proposes a new algorithm for generating the two approximate logics. Through the generating process, the algorithm considers the intrinsic failure probabilities of each gate and utilizes a confidence interval estimate equation to minimize required computations. The technique is applied to two fault models, Stuck-at and SET, and the separate results are compared and discussed. The results show that the technique can reduce the error 75% with an area penalty of 46% on some circuits. The delay overheads of this technique are always two additional layers of logic.
The two proposed SET mitigation techniques are both applicable to generic combinatorial logics and with high flexibility. The simulation shows promising SET mitigation ability. The proposed mitigation techniques provide designers more choices in developing reliable combinatorial logic in ground-level applications.||