Paper Info
Title
Low-power Accuracy-configurable Carry Look-ahead Adder Based on Voltage Overscaling Technique
Abstract
In this paper, a low-power accuracy-configurable block-based Carry Look-ahead Adder (AC-CLA) is proposed. The structure employs the voltage over scaling and number of approximate blocks as the approximation knobs for improving the energy consumption as well as the reliability and lifetime of the adder. While the former may be set in the design time as well as the runtime, the latter may only be invoked in the design time. In this adder, for a given accuracy level, some of the blocks work in the approximate mode by using over-scaled voltages. The block-based structure enables applying the overscaled voltage for each block independently. The efficacy of the adder depends on the number of the approximate blocks as well as the VOS voltage levels used for these blocks. The use of lower VOS voltage levels for the blocks responsible for lower significant bits which have higher switching activities is the key for reducing the power consumption of the adder while having the error within a tolerable limit. The structure requires few level shifters making the realization overhead low. The efficiency of the AC-CLA structure is studied using a 15 nm FinFET technology. The results of the study indicate that in the approximate mode up to 57% energy saving may be achieved. In addition, for this adder, the BTI induced delay degradation of the adder over 10 years decreases by up to 7% compared to 50% in the case of the exact operating mode. Finally, the efficacy of AC-CLA adder is assessed in a neural network for the classification application.
Year
DOI
Venue
2020
10.1109/ISQED48828.2020.9137039
2020 21st International Symposium on Quality Electronic Design (ISQED)
Keywords
DocType
ISSN
Approximate Computing,Lifetime,Aging,FinFET,Accuracy,Reconfigurability,Energy-Efficient
Conference
1948-3287
ISBN
Citations 
PageRank 
978-1-7281-4207-4
0
0.34
References 
Authors
0
3
Name
Order
Citations
PageRank
Hassan Afzali-Kusha173.21
Mehdi Kamal218930.41
Massoud Pedram378011211.32