from which memory-deep it is more meaningfully to use a RAM

I am searching for the border when it is meaningfull to use a RAM or
RegisterFile instead of of a RTL register description.
Has someone empirical values when it is better to use RAM hardmacros
to save area ?

Thanks
calzi

 

[Petter Gustad]

calzinide@yahoo.de writes:

I am searching for the border when it is meaningfull to use a RAM or
RegisterFile instead of of a RTL register description.
Has someone empirical values when it is better to use RAM hardmacros
to save area ?

That depends on the implementation technology, availability, speed of
embedded RAM's and size of your registerfile. Also how many ports you
need to have on your registerfile is very important. If you have 8
speculative execution units reading data from a registerfile then it's
cumbersome to implement in a technology which only has dual-port RAM
(you can of course use 4 RAM's if you have plenty of unused RAM).

Petter

--
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

 

Thank's
shure it depends on the technology , but i am searching for nosefaktor
when it is meaning full for simple RAM implementations. for example: 8
bit addr 16 data etc. just a rule for design guidlines not for a stable
technology dependent design.

Petter Gustad schrieb:

calzinide@yahoo.de writes:

I am searching for the border when it is meaningfull to use a RAM or
RegisterFile instead of of a RTL register description.
Has someone empirical values when it is better to use RAM hardmacros
to save area ?

That depends on the implementation technology, availability, speed of
embedded RAM's and size of your registerfile. Also how many ports you
need to have on your registerfile is very important. If you have 8
speculative execution units reading data from a registerfile then it's
cumbersome to implement in a technology which only has dual-port RAM
(you can of course use 4 RAM's if you have plenty of unused RAM).

Petter

--
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

 

[J o h n _ E a t o n (at)]

calzinide@yahoo.de wrote:

I am searching for the border when it is meaningfull to use a RAM or
RegisterFile instead of of a RTL register description.
Has someone empirical values when it is better to use RAM hardmacros
to save area ?

Thanks
calzi

For asics the threshold is currently about 128-256 bits. Below that you

create a rtl description. Above that you instantiate a sram hard macro.

In either case if your logic can fit into a ram form-factor then it is
a good idea to code it as sram and port all the connections out of your
module. You can decide later to use either a hard macro or simply drop
in a synthesized rtl description. It's easier to design for a hard macro
upfront and not use it than it is to go in later and convert a rtl design
into a sram.

John Eaton

 

[Kim Enkovaara]

J o h n _ E a t o n (at) hp . com (no spaces) wrote:

For asics the threshold is currently about 128-256 bits. Below that you
create a rtl description. Above that you instantiate a sram hard macro.

That also depends on the technology and memory compilers. Some compilers
support memories made out of latches and 3-state busses, and those are used
between register files and real srams. Those latch structures can be tested
with scan chains, sram requires bist which takes some space also.

In either case if your logic can fit into a ram form-factor then it is
a good idea to code it as sram and port all the connections out of your
module. You can decide later to use either a hard macro or simply drop
in a synthesized rtl description. It's easier to design for a hard macro
upfront and not use it than it is to go in later and convert a rtl design
into a sram.

I agree with this. It also easier to port the code to other technologies etc.
when the rams are well separated from the code with wrappers.

--Kim