Quantifying academic placer performance on custom designs

摘要: There have been significant prior efforts to quantify performance of academic placement algorithms, primarily by creating artificial test cases that attempt mimic real designs, such as the PEKO benchmark containing known optimas [5]. The idea was create benchmarks with a optimal solution and then measure how far existing placers were from optimal. Since do not necessarily correspond properties VLSI netlists, conclusions met some skepticism. This work presents two custom constructed datapath designs perform common logic functions hand-designed layouts for each. new generation is compared against them see performed these design styles. Experiments show all wirelengths significantly greater manual solution; solutions range 1.75 4.88 times wirelengths. These testcases will be released publically stimulate research into automatically solving structured problems.