A methodology for performance driven incremental placement with high level exploration
M. Mukherjee, Ranga Vemuri
Abstract
M. Mukherjee, Ranga Vemuri
Abstract
In order to accurately evaluate a design decision made during high level synthesis (HLS), it is essential to clearly account for its effects on the final layout. We propose a mixed integer linear programming (MILP) based approach to perform incremental global placement driven by delay budgeting in a HLS framework. The methodology incrementally updates global placement to accommodate design decisions made during HLS. Given a target clock speed, a delay budget is associated with interconnects that change during exploration. The placer aims at accommodating changes in the RTL netlist while meeting design constraints. Integration of this strategy during HLS design space exploration allow the designer to examine various RTL structures for better post-layout physical characteristics. Experiments show that the approach is extremely fast and yields compact placement results for accommodating HLS design decisions.
OpenAlex reports 1 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
In order to accurately evaluate a design decision made during high level synthesis (HLS), it is essential to clearly account for its effects on the final layout. We propose a mixed integer linear programming (MILP) based approach to perform incremental global placement driven by delay budgeting in a HLS framework. The methodology incrementally updates global placement to accommodate design decisions made during HLS. Given a target clock speed, a delay budget is associated with interconnects that change during exploration. The placer aims at accommodating changes in the RTL netlist while meeting design constraints. Integration of this strategy during HLS design space exploration allow the designer to examine various RTL structures for better post-layout physical characteristics. Experiments show that the approach is extremely fast and yields compact placement results for accommodating HLS design decisions.
Key concepts: Netlist, Design space exploration, High-level synthesis, Computer science, Placement, Integer programming, Physical design, Place and route