Synthesis of FSMs..

[VHDL User]

Hi All,
Suppose I have a description of a FSM in which I use self defined (enum)
data type State which can be amongst {"Red","Black","Blue"}.
Clearly 2 FFs will suffice to describe 3 states.My question is related
to the synthesis of such a machine:
1.How does the tool decide the encoding mechanism? I can have 00,01,11 or
any 3 of 4 options.Is there a way to fix this apriori within VHDL Code
itself ? How do I otherwise specify my preferences? ( I use Synplicity.)
2.When would a encoding of 3 states in a 001,010,100 manner be more
"useful" or "better",if at all? Clearly,more FFs would be needed,but
would this affect the speed ?
Thanks a lot,
Bye

 

[David R Brooks]

VHDL User <supreet@wrongdomain.com> wrote:

:Hi All,
: Suppose I have a description of a FSM in which I use self defined (enum)
:data type State which can be amongst {"Red","Black","Blue"}.
: Clearly 2 FFs will suffice to describe 3 states.My question is related
:to the synthesis of such a machine:
:1.How does the tool decide the encoding mechanism? I can have 00,01,11 or
:any 3 of 4 options.

A straightforward, general way is like this:

subtype TSTATE is std_logic_vector(1 downto 0);

constant RED : TSTATE := "00";
constant BLACK : TSTATE := "01";
constant BLUE : TSTATE := "10";

signal STATE_VARIABLE : TSTATE;

When you use a case..when to select your state from this, assign the
"others" case to an illegal-state catcher (eg reset).

:Is there a way to fix this apriori within VHDL Code
:itself ? How do I otherwise specify my preferences? ( I use Synplicity.)

This is called "one hot" encoding. Where to use it is a tradeoff: you
have more flipflops, but less decoding logic. Typically, one-hots run
faster than fully decoded states (less logic).
If you use one-hot, you may need to include safety logic to catch
cases of two flipflops getting set together (no, it shouldn't happen).

:2.When would a encoding of 3 states in a 001,010,100 manner be more
:"useful" or "better",if at all? Clearly,more FFs would be needed,but
:would this affect the speed ?
:Thanks a lot,
:Bye

 

[anupam]

Hi Supreet,

1.
If u want to encode in your manner then u have to declare the state as
constant like as
constant red : std_logic_vector(1 downto 0) := "00"
.....
....
2.
for this you have to assign the state machine in one - hot encoding
method.Definitely the FFs used will be more and will consume more logic
but the speed of design will be improved (increased).

Hope it is helpfull to u .

Regards,
Anupam Garg

 

[Niv]

One thing we do where I work is to define 2^N states, so that 2^N is always
greater than the number of states we need,
so in your case there would be a fourth "spare" state, say "sp_01".

We would then write the state machine so that it passed through all
spare states after reset before reaching the "idle" or "start" state.

We would use binary encoding.

This forces the synth to produce only N FF's, with all possible states
declared and used, albeit fleetingly for the spare states.

This helps with safety critical systems; and testbenches can test that all
states are exercised.

Yes, the design will be slower than one-hot (a bit, size dependant I
suppose).

Hope that helps, Niv.

 

[cristian]

"Niv" <niv.nospam.goaway@ntlworld.com> wrote in message news:<lY95d.238$fy3.222@newsfe6-win.ntli.net>...

One thing we do where I work is to define 2^N states, so that 2^N is always
greater than the number of states we need,
so in your case there would be a fourth "spare" state, say "sp_01".

We would then write the state machine so that it passed through all
spare states after reset before reaching the "idle" or "start" state.

We would use binary encoding.

This forces the synth to produce only N FF's, with all possible states
declared and used, albeit fleetingly for the spare states.

This helps with safety critical systems; and testbenches can test that all
states are exercised.

Yes, the design will be slower than one-hot (a bit, size dependant I
suppose).

Hope that helps, Niv.

Dear VHDL_User,

FSM Synthesis is a complicated subject. To simplify the problem, two
main subjecst: which architecture you will use to implement your
design, FPGA or CPLD. Then, which synthesis tool you will use to
synthesize your code. I am familiar with both Synplify and Leonardo.
For the number of states you have in your FSM, 3 or 4, there are no
too many options, both synthesis will use sequential as default, for
more than 5 states they will use one-hot even though is not the best
solution for CPLDs. If you want to overwrite the default sequantial
coding style you will need to use synthesis attributes. As we all know
this is not good if we are looking for portable designs. Depending on
the synthesis tool will be the attribute to use. Check the your
synthesis tool's user guide to find which attribute you need to use.

best regards,

cristian