ROM vs "Combination lookup table"

[aaronhe]

Our project needs some high speed ROM, but the ROM from many verndors can't meet our spec. So I try to make it with "Combination lookup table", just use standard cell.

The results show that area and speed of "Combination lookup table" is better than ROM.

Could anyone tell me some shortcoming of my method?

 

[rca]

none!
we used the same way.
the main issue you can not change the table easily, be sure to have check the table two times.

 

[srizbf]

power is also reduced using combi LUt . am i right ?

srizbf
8thjuly2010

 

[randyest]

power is also reduced using combi LUt . am i right ?

That depends on the process. In my experience ROM is smaller and lower power than gates.

(P.S. I'm just wondering why you put your name and date at the bottom of your posts?)

edit: \/\/\/\/ Thanks for the reply. It's not a problem, of course, I was just wondering if it had some special significance

 

[srizbf]

it is a coincidence that in todays posts only i thought of removing dates .
my previous posts in other forum(today) i have adapted it.

now after reading your post , that name is also redundant.

anyway thanks a lot for pointing that.

(oh the habit ...)

 

[RBB]

A combinatorial LUT could take all layers to change its programming. A ROM should ideally only take one layer to change its programming.

 

[supercat]

The results show that area and speed of "Combination lookup table" is better than ROM.

Could anyone tell me some shortcoming of my method?

A ROM provides a means of representing any combination of 1's and 0's in a predictable amount of space, with predictable timing. The performance and timing will be essentially independent of the data being represented.

Using combinatorial logic which is optimized for a particular arrangement of 1's and 0's will allow more efficient circuitry for some arrangements of 1's and 0's, but for other arrangements it will be no better than a ROM.

As a simple example, suppose a 256x8 ROM holds the values 255-0 in that order. The ROM could easily be replaced by eight inverters. On the other hand, suppose the ROM holds bits 1-8 of the first 256 odd prime numbers. What collection of logic would you use for that?

 

[aaronhe]

power is also reduced using combi LUt . am i right ?

srizbf
8thjuly2010

No, ROM's dynamic power and leakage power are both lesser than "Combination lookup table"

Added after 5 minutes:

The results show that area and speed of "Combination lookup table" is better than ROM.

Could anyone tell me some shortcoming of my method?

A ROM provides a means of representing any combination of 1's and 0's in a predictable amount of space, with predictable timing. The performance and timing will be essentially independent of the data being represented.

Using combinatorial logic which is optimized for a particular arrangement of 1's and 0's will allow more efficient circuitry for some arrangements of 1's and 0's, but for other arrangements it will be no better than a ROM.

As a simple example, suppose a 256x8 ROM holds the values 255-0 in that order. The ROM could easily be replaced by eight inverters. On the other hand, suppose the ROM holds bits 1-8 of the first 256 odd prime numbers. What collection of logic would you use for that?

FFT usage, the data is not so simple as 255-0

 

[Karandeep]

A combinatorial LUT could take all layers to change its programming. A ROM should ideally only take one layer to change its programming.

Q1: What is the difference between both the LUT ? What kind of layers are being discussed here ? Can anyone please elaborate.

Q2: Please also let me know the fact that the CONSTANTS we declare and use in the RTL design are ROM LUT or Combinational LUT ?

Please enlighten