|
Sofia Benchmark
Benchmarking a library of standard cells
Assessing the comparative performances of several libraries in a reliable way is a tricky project as it deals with statistical issues…Two main techniques are traditionally used in the industry to do so.
The so-called « cell-by-cell » approach.
It consists in taking one or two basic cells like a nand and a flip-flop and compare their area, dynamic power consumption, leakage and speed.
The obvious drawbacks of this method are that it only assesses a few cells, which are not necessarily representative. It is even less relevant for a comparison of libraries with different structures like a traditional library and a reduced library.
The second method is based on a Synthesis and Place and Route on a block representative of some SoC
This method is the most reliable for the specific SoC, but there are two major drawbacks: it is time consuming and the conclusions are not relevant for other SoCs.
Save time with Dolphin’s « Sofia benchmark »
In order to compare quickly but correctly libraries with different structures like reduced libraries and traditional libraries, a benchmark of a new kind is now available.
The main benefits of the Sofia benchmark are:
- a straight and simple process: it takes around 15 minutes to get the comparison results!
- representativity: the Sofia benchmark is using 6 cells representative of the typical paths in a majority of logic circuits.
Each cell is weighted depending on the percentage that it represents in the path, issued from on a large sample of circuits. These weights vary depending on if it is a traditional library or a reduced library like SESAME.
How to use it?
- For each of the 4 criteria: area, dynamic power, leakage, speed, a table with the 6 weighted cells is provided. See example below for area:
Area |
the value is defined as the area of the smallest cell of each type |
|
DFC 3 |
ND 21 |
ANR 2 |
MX 22 |
ADD 2 |
IN 01 |
FoM area |
Library |
Value |
Weight |
Value |
Weight |
Value |
Weight |
Value |
Weight |
Value |
Weight |
Value |
Weight |
|
SESAME |
|
6,60% |
|
26,70% |
|
16,70% |
|
10% |
|
6,70% |
|
33,30% |
0 |
Customer's library |
|
10% |
|
15% |
|
25% |
|
15% |
|
10% |
|
25% |
0 |
- Fill in the area values for each cell of your library. These values can be found in the .lib file and are prefilled-in for Dolphin’s library
- Once each value is entered in the table, you will get the “Figure of Merit” (last column “FoM” in the example) of each library on the selected criteria
- Repeat the same steps for the other criterions: dynamic power, leakage, speed
- The global Figure of Merit, representing the overall performance of the library, is based on the multiplication of the Figure of Merit of each criterion (area, dynamic power, leakage, speed), giving the same importance to each of the main criterion. The importance of each criterion can be weighted to reflect the chosen optimization of the SoC.
|
|
Printer-friendly version
