- new switches
- use constant delay model
- worst-case intrinsic delays
- based on fanout-3
- modify VPR to model fanout delay
- measure critical path delay,
fanout delay:
- old buf: +5% avg, 16% worst-case
- new bufm: +1% avg, +5% worst-case
- bufm better than bufp
|
Normalized Results 4-input LUT, 6 LUTs per cluster
Buffer Type |
Area |
Delay |
Area*Delay |
buf |
1.0 |
1.0 |
1.0 |
bufp |
0.97 |
0.98 |
0.95 |
bufm |
0.97 |
0.93 |
0.91 |
|