Adaptive Fault-Tolerant Architecture for Unreliable Technologies With Heterogeneous Variability

Aymerich, Nivard; Cotofana, Sorin; Rubio, Antonio

doi:10.1109/tnano.2012.2199513

Cited by 5 publications

(8 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In contrast, the AD-AVG cell architecture does not assume homogeneity and thus evaluates the variability at each input, by employing a Variability Monitor block, and calculates the set of different weights to optimize reliability. In [11], it has been demonstrated that the AD-AVG cell can efficiently tolerate high amounts of heterogeneous variability and accumulated degradation in the physical replicas.…”

Section: Ad-avg Cellmentioning

confidence: 99%

“…The analytical proof of the aforementioned equation can be found in [11]. The input variances σ 2 i are estimated by means of the Variability Monitor block that is subject to noise…”

Section: Ad-avg Cellmentioning

confidence: 99%

“…The averaging weights are reconfigured by the weight drivers according to the input variance estimators and the optimal averaging weights' formula, see (5). A detailed presentation and discussion about this calculation can also be found in [11].…”

Section: Ad-avg Cellmentioning

confidence: 99%

“…However, this condition is no longer valid for the current technologies as heterogeneity starts playing a relevant role. To optimize the AVG architecture in nonhomogeneous variation environments, we proposed the adaptive averaging (AD-AVG) architecture [11]. This enhanced AVG technique is capable of tolerating nonhomogeneous input variations and the effects of degradation by cleverly adapting the input weight values.…”

mentioning

confidence: 99%

“…The basic AD-AVG principle is to assign larger weights to the most reliable inputs, based on the measure of the associated variability, and smaller weights to the ones more prone to be unreliable. We have analyzed in detail the AD-AVG capabilities in [11], and we have recently discovered an interesting but counter-intuitive phenomenon occurring in AD-AVG structures, the so-called degradation stochastic resonance (DSR) [12], that deserves further study.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Controlled Degradation Stochastic Resonance in Adaptive Averaging Cell-Based Architectures

Aymerich

Cotofana

Rubio

2013

IEEE Trans. Nanotechnology

Self Cite

View full text Add to dashboard Cite

In this paper, we first analyze the degradation stochastic resonance (DSR) effect in the context of adaptive averaging (AD-AVG) architectures. The AD-AVG is the adaptive version of the well-known AVG architecture . It is an optimized fault-tolerant design for future technologies with very high rates of failures and defects. With system degradation the AD-AVG reliability is diminishing, as expected, but at a certain moment in time it increases due to the DSR occurrence, which is counterintuitive. We study this phenomenon under various redundancy levels and noise condition. If we take for example a 20-input AD-AVG with particular noise conditions, our simulations indicate an initial yield decrease from 1 to 0.89 with the system degradation, then a grow up to 0.94 at the DSR peak, and finally a decrease to zero when the system is reaching its end of life. Subsequently, we introduce a method to induce DSR in an AD-AVG structure, regardless of the degradation level, when this results in reliability improvement. To achieve this, we augment the AD-AVG with per input controllable noise injectors that can be utilized to induce virtual circuit degradation and create the required conditions for the DSR peak appearance. With this scheme the beneficial DSR effect is created even though the actual DSR system degradation (aging conditions) is not reached. This allows us to provide an optimum and nearly flat reliability level at any time before the DSR peak degradation level. Our experiments suggest that when we apply this method to the same 20-input AD-AVG, we obtain a guaranteed yield level of 0.94 from fresh devices to the DSR peak degradation level with a maximum yield of 0.97. In this way, a minimum yield level can be guaranteed, by determining at design time the required AD-AVG redundancy that provides it, for the entire life of the system.

show abstract

Section: Ad-avg Cellmentioning

confidence: 99%

“…The analytical proof of the aforementioned equation can be found in [11]. The input variances σ 2 i are estimated by means of the Variability Monitor block that is subject to noise…”

Section: Ad-avg Cellmentioning

confidence: 99%