Efficient FPGA Routing Architecture Exploration Based on Two-Stage MUXes

Jide Zhang,Kaixiang Zhu,Kaichuang Shi,Lingli Wang,Hao Zhou
DOI: https://doi.org/10.1109/asicon58565.2023.10395964
2023-01-01
Abstract:Employing large routing multiplexers (MUXes) in FPGA results in significant area and delay overheads. Hence, the two-stage cascaded structure with small MUXes can reduce the area and delay effectively. However, the manual design of a two-stage MUX topology is challenging to find optimal architecture. An automatic switch pattern exploration framework is proposed based on a novel routing architecture, TSRB (Two-Stage MUX Routing Block), which consists of two-stage MUXes in switch box (SB) and local interconnection block (LIB). The framework can enable the router to select commonly used switches based on the usage of switch connections under the premise of solving congestion, leading to a minimal set of switch types used in the architecture. The TSRB switch pattern is generated aided by a pre-exploration method to avoid manual design. The pattern optimization is then followed by two-stage MUX exploration. Our exploration framework can generate a TSRB switch pattern and optimize the pattern by pruning more than half of the switch connections, achieving an average 7% shorter critical path delay with a 5% area reduction compared to the one-stage large MUX routing architecture.
What problem does this paper attempt to address?