High-performance, Cost-effective 3D Stacked Wide-Operand Adders

Journal Article (2016)
Author(s)

George Voicu (TU Delft - Computer Engineering)

S.D. Cotofana (TU Delft - Computer Engineering)

Research Group
Computer Engineering
Copyright
© 2016 G.R. Voicu, S.D. Cotofana
DOI related publication
https://doi.org/10.1109/TETC.2016.2598290
More Info
expand_more
Publication Year
2016
Language
English
Copyright
© 2016 G.R. Voicu, S.D. Cotofana
Research Group
Computer Engineering
Bibliographical Note
Accepted Author Manuscript@en
Issue number
2
Volume number
5
Pages (from-to)
179-192
Reuse Rights

Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.

Abstract

Through-Silicon Vias (TSV) based 3D Stacked IC (3D-SIC) technology introduces new design opportunities for wide operand width addition units. Different from state of the art direct folding proposals we introduce two cost-effective 3D Stacked Hybrid Adders with identical tier structure, which potentially makes the manufacturing of hardware wide-operand fast adders a reality. An N-bit
adder implemented on a K identical tier stacked IC performs in parallel two N=K-bit additions on each tier according to the anticipated computation principle. Inter-tier carry signals performing the appropriate sum selection are propagated by TSVs. The practical implications of direct folding and of our hybrid carry-select/prefix approaches are evaluated by a thorough case study on 65nm CMOS 3D adder implementations, for operand sizes up to 4096 bits and 16 tiers. Our simulations indicate that in almost all configurations at least one of the two proposed 3D stacked hybrid approaches is faster than the fastest 3D folding approach. When considering an appropriate metric for 3D designs, i.e., the delay-footprint-heterogeneity product, the hybrid adders substantially outperform the folding counterparts by a factor in-between 1:67 and 23:95.