大家如果有空 也看看吧 Repeaters are used to reduce the delay on global wires. As feature size scaling down, we need more repeaters to meet the timing requirement. Such huge number of repeaters will occupy a lot of routing resources and make the routing more difficult. This paper proposes a way to reduce the count of repeaters by minimizing the number of global nets. The technique is to add an on-line wirelength estimation, and use this information to guide the top-down placement. This concept is simple but it has a good result on reducing global nets. However, several points need to be taken care of: 1: CL (critical wirelength) is assumed to be 20%, 30%, 40% of the chip's half perimeter. A more realistic calculation of CL (ex: using the parameters of certain technology) can give us a better insight into this problem. 2: LBB and MLBB are little used to estimate the wirelength, and the results are much worse than CBB and MCBB. Why the author spent a lot of time to compare the results between LBB (LBB) with CBB (MCBB)? 3: In table 2 and 3, the term “HPWL” and “WireLength” are a little vague. One may consider them to be total wirelength, while another may think it as the sum of wirelength of the wires longer than CL. The term “# Nets” also needs more explicit explanation. 4: In page 4, the author claims that a larger alpha tends to increase the total wirelength. We look forward to more explanation about it. 5: Formula 2 in page4: “W”CBB->”M”CBB --

※ 發信站: 批踢踢實業坊(ptt.cc) ◆ From: 140.112.48.60