Result: A New Basic Logic Structure for Data-Path Computation

Title:
A New Basic Logic Structure for Data-Path Computation
Publication Year:
2014
Collection:
Ecole Polytechnique Fédérale Lausanne (EPFL): Infoscience
Document Type:
Conference conference object
Language:
unknown
ISBN:
978-1-4503-2671-1
1-4503-2671-4
Relation:
https://infoscience.epfl.ch/record/201910/files/PEG_FPGA14.pdf; Proceedings of the 22nd ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA 2014); 22nd ACM/SIGDA International Symposium on Field-Programmable Gate Arrays (FPGA 2014); https://infoscience.epfl.ch/handle/20.500.14299/107149
DOI:
10.1145/2554688.2554701
Accession Number:
edsbas.EF9F4CF4
Database:
BASE

Further Information

Nowadays, Field Programmable Gate Arrays (FPGA) implement arithmetic functions using specific circuits at the logic block level, such as the carry paths, or at the structure level adopting Digital Signal Processing (DSP) blocks. Nevertheless, all these approaches, introduced to ease the realization of specific functions, are lacking of generality. In this paper, we introduce a new logic block that natively realizes arithmetic functions while preserving the versatility to implement general logic functions. It consists of a partially interconnected matrix of signal routers driven by comparators. We demonstrate that this structure can realize (i) any 2-output 2-input logic function or (ii) any single-output 3-input logic function or (iii) specific logic, such as arithmetic functions, with up to 4-output and 8-inputs. As compared to a standard 6-input Look Up Table (LUT), the proposed block requires roughly the same area but is 35.3% faster. Even though the proposed block has not the same exhaustive configurability of a 6-input LUT, there are arithmetic functions realizable in a single block that do not fit in one, or even more, 6-input LUT. For example, a single block inherently implements an entire 3-bit adder that requires 3× more resources with LUTs plus also custom circuitry. From a system level perspective, we show that a 256-bit adder is implemented with a gain on area×delay product of 31% as compared to its traditional LUT-based counterpart. ; LSI1