Difference Logic Cells <=> Slices

Michiel,

The issue here is what marketing does after we IC designers are done.

With good intentions, they want to promote the product, and have it appear competitive in every sense. They also want to be honest, and every claim is backed up by real data.

So, if there is one more mux, one more carry input, anything at all that gets added to a LUT+stuff+DFF, they "scale" the "effectiveness" of the "logic cell."

To me, I count LUTs and DFFs. With the introduction of V5, I do not count a 6-LUT as some number of 4-LUTs, it is still a 6-LUT.

Unfortunately, 6-LUTs are used from hardly at all to as much as 58% of a design, meaning the 6-LUT on average is quite an improvement (more logic done in fewer stages, leading to faster, and lower power). This makes it hard to compare a part with 4-LUTs to a part with 6-LUTs.

So, getting back to your question: are you just trying to untangle the data sheet to see the underlying hardware features?

Looking at Table 1 from

we see in note 2: "Logic Cell = 1 4-LUT + 1 DFF + Carry Logic

Page 6, defines a "CLB" as 4 slices.

A Slice is defined as two 4-LUTs (function generators), to FFs, all the arithmetic stuff, large muxes (F5,F6,F7), carry lookahead, horizontal cascade.

Back to table 1, a 2VP2 has 1408 slices, so that makes:

1408 X 2 = 2816 4-LUTs 1408 X 2 = 2816 DFFs etc.

I like to remember that a 'CLB' has 8 LUTs, 8 DFFs (an octet of useful stuff...) in VII and VII Pro. And a CLB is 4 slices, and each slice is

2 4-LUTs and 2 DFFs.

So back to logic cells, the 2VP2 has 3168 logic cells.

It has 2816 identifiable LUT+DFF pairs, so where does that factor of

1.125 come from? That is the extra stuff: arithmetic, large muxes (F5,F6,F7), carry lookahead, horizontal cascade.

For me, I would not bother to dress up the number, but if you synthesize a few thousand designs, the number holds up (can be justified) as an average improvement compared to synthesizing the design with those resources locked out.

Bottom line: this 'inflation factor' is justified by actual data, but it is very confusing (if all you really wanted to know was how many LUTs and flops there are).

Austin

Mich wrote: