Routing experiments to save area
- switch blocks with diversity (universal, Wilton)
- switch blocks with fewer connections
Interconnect details
- only length 4 wires
- 50% tracks use 6x buffers
- 50% tracks use 16x pass transistors
Results
- many experiments, typical results shown here
- baseline disjoint with 3 connections
- almost any perturbation lowers area
- savings usually small (3%)
- just noise? not likely, consistent results
- delay noisy (3%), no trend
- fewer endpoint switches
- increases track count, saves area
- lower area than diverse switch blocks
- combination of diversity and fewer switches (not shown)
- still only 3% area savings
|
Normalized Results
4-input LUT,
6 LUTs per cluster
S Block
Type |
Min Chan.
Width |
Area |
Delay |
| 3 endpoint connections |
| disjoint |
1.0 |
1.0 |
1.0 |
| universal |
0.969 |
0.983 |
1.036 |
| Imran (Wilton) |
0.977 |
0.987 |
1.033 |
| 2 endpoint connections |
| disjoint |
1.012 |
0.966 |
0.982 |
| 1 endpoint connection |
| disjoint |
1.123 |
0.971 |
0.997 |
|