Re: paramsets - statistics

Ilya,
     The x, y, angle, hflip and vflip attributes are provided along with
m and n to allow layout dependent monte carlo. This is a new capability
that is meant to provide more accurate calculations of correlations
based on the actual size and placement of components.

matched and correlation were considered but were removed. matched refers
to particular instances, and so did not fit in the paramset paradigm,
and correlation was ill-defined for distributions other than gaussian.

In the current proposal, correlation is implemented by the user by
combining random variables into new variables, and the matched
functionality can be implemented by with either the x, y, etc.
attributes, or by creating additional paramsets.

-Ken

Ilya Yusim wrote:
> Geoffrey,
>
> 1) Statistics:
> It would be good to get the statistics into the proposal. I don't
> know what the restrictions are for OOMR's but, the statistics block is a
> special type of module (with only statistical parameters). Maybe we can
> allow OOMR's into the statistics block.
>
> In addition to keywords statistics, process, and mismatch, we would
> need: matched and correlate.
>
> 2) Hierarchical attributes: m and n, are important. Other's do not seem
> to be. I have never seen them in side a device. Should they be
> restricted to paramsets? Where are they typically used? Note, only n
> and m have clear rules that can be automatically applied. For others,
> the rules are process dependent.
>
> Ilya
>
>
>
> Geoffrey.Coram wrote:
>
>> I was reviewing the paramsets proposal to see if the
>> excerpt I sent out recently covered all the aspects.
>> It does not: sections 2.9 - 2.12 are missing.
>>
>> 2.9 and 2.10 deal with Monte-Carlo and statistics,
>> and require the addition of three keywords,
>> statistics, process, and mismatch. Also, the
>> statistical values are stored in variables and
>> accessed through out-of-module references. I can't
>> find OOMRs documented; I'd heard that they cannot
>> be used to access variables.
>>
>> 2.11 and 2.12 deal with hierarchical attributes
>> and layout descriptions (m,n,x,y,angle used for
>> mismatch). We're still waiting for a revised
>> proposal on mfactor.
>>
>> In the conference call on June 1, we decided that
>> the LRM for CM was not complete without paramsets
>> and mfactor. My question is: with mfactor but
>> without the items above, is this a complete LRM?
>> Is it at all likely that we could get the rest
>> of the items approved?
>>
>> My sense is that verilog-a is not useable for CM
>> without a .model card; we need the basic functionality.
>> I think the other features would be nice, but they
>> haven't been thought through as much, and they could
>> go in a future revision of the LRM.
>>
>> -Geoffrey
>>
>>
>
>
>
Received on Fri Jun 4 17:26:28 2004