2002•Unpublished venueRequires access

Placement driven retiming with a coupled edge timing model

Ingmar Neumann, Wolfgang Kunz

Open publisher page 5 citations

Abstract

Retiming is a widely investigated technique for performance optimization. It performs powerful modifications on a circuit netlist. However, often it is not clear whether the predicted performance improvement will still be valid after placement has been performed. This paper presents a new retiming algorithm using a highly accurate timing model taking into account the effect of retiming on capacitive loads of single wires as well as fanout systems. We propose the integration of retiming into a timing-driven standard cell placement environment based on simulated annealing. Retiming is used as an optimization technique throughout the whole placement process. The experimental results show the benefit of the proposed approach. In comparison with the conventional design flow based on standard FEAS (Leiserson and Saxe, J. VLSI and Computer Sys., pp. 41-67, 1983; and Algorithmica vol. 6, no 1, pp. 5-35, 1991.), our approach achieved an improvement in cycle time of up to 34% and 17% on the average.

About this research paper

What this paper is about

Retiming is a widely investigated technique for performance optimization. It performs powerful modifications on a circuit netlist. However, often it is not clear whether the predicted performance improvement will still be valid after placement has been performed. This paper presents a new retiming algorithm using a highly accurate timing model taking into account the effect of retiming on capacitive loads of single wires as well as fanout systems. We propose the integration of retiming into a timing-driven standard cell placement environment based on simulated annealing. Retiming is used as an optimization technique throughout the whole placement process. The experimental results show the benefit of the proposed approach. In comparison with the conventional design flow based on standard FEAS (Leiserson and Saxe, J. VLSI and Computer Sys., pp. 41-67, 1983; and Algorithmica vol. 6, no 1, pp. 5-35, 1991.), our approach achieved an improvement in cycle time of up to 34% and 17% on the average.

Why it matters

OpenAlex reports 5 citations for this work. Citation counts describe recorded attention and do not establish research quality.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

Retiming is a widely investigated technique for performance optimization. It performs powerful modifications on a circuit netlist. However, often it is not clear whether the predicted performance improvement will still be valid after placement has been performed. This paper presents a new retiming algorithm using a highly accurate timing model taking into account the effect of retiming on capacitive loads of single wires as well as fanout systems. We propose the integration of retiming into a timing-driven standard cell placement environment based on simulated annealing. Retiming is used as an optimization technique throughout the whole placement process. The experimental results show the benefit of the proposed approach. In comparison with the conventional design flow based on standard FEAS (Leiserson and Saxe, J. VLSI and Computer Sys., pp. 41-67, 1983; and Algorithmica vol. 6, no 1, pp. 5-35, 1991.), our approach achieved an improvement in cycle time of up to 34% and 17% on the average.

Key concepts: Retiming, Netlist, Computer science, Very-large-scale integration, Simulated annealing, Standard cell, Parallel computing, Floorplan

Related papers

Back to paper searchBrowse research topicsOriginal source
Placement driven retiming with a coupled edge timing model — Research Paper | ScholarLens