TY - GEN
T1 - Design of a 3-D stacked floating-point adder
AU - Tada, Jubee
AU - Egawa, Ryusuke
AU - Kobayashi, Hiroaki
PY - 2013/12/1
Y1 - 2013/12/1
N2 - Three-dimensional (3-D) stacked integrated circuit (SIC) technologies have been expected to overcome the limitation in the design of microprocessors integrated by two-dimensional (2-D) implementations. 3-D SIC technologies enable to stack multiple integrated silicon layers. In the design of 3-D stacked arithmetic units, the circuits are partitioned into several subcircuits, and each sub-circuit is placed on one layer. In order to exploit the potential of the 3-D SIC, a sophisticated partitioning should be required. In this paper, four partitioning patterns for a 3-D stacked floating-point adder are proposed, which are based on two basic ideas. One idea focuses on the structure of a 2-path floating point adder, and the other idea focuses on the large barrel shifters. Four implementations of a 3-D stacked double-precision floating-point adder are designed based on these partitioning patterns and evaluated. Experimental results show that the 3D stacked double precision floating-point adder implemented on four layers achieves up to a 16.4% delay reduction compared to the 2-D implementation.
AB - Three-dimensional (3-D) stacked integrated circuit (SIC) technologies have been expected to overcome the limitation in the design of microprocessors integrated by two-dimensional (2-D) implementations. 3-D SIC technologies enable to stack multiple integrated silicon layers. In the design of 3-D stacked arithmetic units, the circuits are partitioned into several subcircuits, and each sub-circuit is placed on one layer. In order to exploit the potential of the 3-D SIC, a sophisticated partitioning should be required. In this paper, four partitioning patterns for a 3-D stacked floating-point adder are proposed, which are based on two basic ideas. One idea focuses on the structure of a 2-path floating point adder, and the other idea focuses on the large barrel shifters. Four implementations of a 3-D stacked double-precision floating-point adder are designed based on these partitioning patterns and evaluated. Experimental results show that the 3D stacked double precision floating-point adder implemented on four layers achieves up to a 16.4% delay reduction compared to the 2-D implementation.
UR - http://www.scopus.com/inward/record.url?scp=84893975377&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84893975377&partnerID=8YFLogxK
U2 - 10.1109/3DIC.2013.6702390
DO - 10.1109/3DIC.2013.6702390
M3 - Conference contribution
AN - SCOPUS:84893975377
SN - 9781467364843
T3 - 2013 IEEE International 3D Systems Integration Conference, 3DIC 2013
BT - 2013 IEEE International 3D Systems Integration Conference, 3DIC 2013
T2 - 2013 IEEE International 3D Systems Integration Conference, 3DIC 2013
Y2 - 2 October 2013 through 4 October 2013
ER -