Hybrid Topologies for Reconfigurable Matrices Based on Nano-Grain Cells

Authors:

Abstract
Nano-grain reconfigurable cells have the potential to replace memory-consuming LUT (Look-Up Table). However, the cells offering the highest area improvement are also those offering the lowest flexibility, i.e. not all the Boolean functions are available. Reaching the same flexibility of LUT is mandatory to reuse existing FPGA tool flows, which can be obtained by clustering cells in a matrix-like architecture. However, while several baseline topologies have been reported, none of them allows to optimally interconnect cells to implement the missing functions. In this paper, we propose a methodology to design hybrid topologies allowing efficient mapping of any function on nano-grain cells based architectures, i.e. including the functions not intrinsically available. The mapping results of benchmark applications on a hybrid topology allow a 36% reduction in the number of transistors over a mapping using a baseline topology.

This publication has 22 references indexed in Scilit: