[SI-LIST] Re: SPICE inaccuracies (was Re: SI models at MGH speeds)
- From: Yu Liu <yu_liu@xxxxxxxxxxxxx>
- To: arpad.muranyi@xxxxxxxxx
- Date: Fri, 27 May 2005 22:47:42 -0700
Arpad,
Personally, I view SPICE as a nonlinear, time variant, ordinary
differential equation
solver. The techniques used in SPICE are mainly numerical integration
(to discretize the
time variable), and numerical iteration (to linearize nonlinear
devices). Thus, from a
solver point of view, the inaccuracies could come from tolerance setting
(how tight
you want to set a convergence criteria during iteration), and time point
selection (how
small you want to choose a time step to catch the changing waveforms).
Fortunately,
in most SPICE tools, the above two are normally set automatically so
that users don't
need to worry them too much (there are also quite some SPICE books
discussing these
techniques and how to tune them).
Of course, SPICE comes with built-in semiconductor device models (diode,
bjt, mosfet, etc).
It seems this is the major inaccuracy source we are focusing on.
Strictly speaking, a semiconductor
device is governed by Poisson's equation and continuity equation, which
are both partial
differential equations. How to incorporate these equations into SPICE is
a formidable task: not
only you need to keep the basic physical behavior of the model, but also
simplify the model significantly
so that it can be evaluated efficiently. To achieve this conflicting
goal, many heuristics are developed
and employed to make the model valid across a wide range of technologies
and geometries. There are
many good examples of such model development, the BSIM1-4 model family
is a good case.
Even after a set of model equations has been developed and verified, it
is not over - you need to
characterize the model to fit into a specific technology, the process of
model parameter extraction.
Nowadays a device model can easily have over 100 model parameters, how
to extract them deserves
a separate topic.
Muranyi, Arpad wrote:
>Hassan,
>
>There may be several reasons for inaccuracies in SPICE.
>
>1) Bad process file. Many times the process files are developed
> for the smallest transistor size, and at other sizes (such as
> in the larger I/O buffers) they are all over the place.
>
Not a single set of model equations can fit all technologies. Likewise,
not a single set of model parameters
can fit all device sizes. That's why a model binning is necessary.
Nowadays, almost all foundries distribute
their device models in binning. It is a surprise to see people use one
set of model parameters for all device sizes.
>
>2) Larger companies have their own proprietary SPICE tools,
> which usually have process parameters which are incompatible
> with commercial tools. Converting these process files to
> something that commercial tools can do results in loss of
> accuracy. Kind of like if you wanted to convert a level=3D13
> MOSFET model to level=3D3.
>
It is not a good idea to convert one model to another. It could be true
that a proprietary
model may better fit to its own technology. But if it can not be used to
the pulic (especially
to its customers), the accuracy advantage can not be fully utilized.
>
>3) I also hear a lot lately that the small geometries we are
> using these days are starting to push the SPICE equations
> to the edge where they are not all that accurate or valid
> any more.
>
I agree, developing model is an on-going effort to take more physical
effects (gate tunneling,
carrier velocity overshoot, to name a few) when the geometry further
shinks. Comparing
to the old models 10-20 years ago, the latest device model can fit
today's dominant
technology amazingly well.
>
>4) For various reasons vendors release reduced "equivalent"
> circuits to the customer, which is not as accurate as the
> full design.
>
Developing a useful model is a delicate balance in generality, accuracy,
complexity, and efficiency,
and demands intimate knowledge of semiconductor physics, process,
circuits, and simulation techniques.
No matter what format it uses (C, IBIS, AMS), a good model always
requires lots of effort and
dedication (and lots of resources).
Best regards,
Yu
==
>
>
>There may be more, but I could only remember these right now...
>
>Arpad
>-----------------------------------------------------------------
>
>
>-----Original Message-----
>From: si-list-bounce@xxxxxxxxxxxxx [mailto:si-list-bounce@xxxxxxxxxxxxx] =
>On Behalf Of Hassan O. Ali
>Sent: Thursday, May 26, 2005 2:17 PM
>To: si-list@xxxxxxxxxxxxx
>Subject: [SI-LIST] Re: SI models at MGH speeds
>
>
>On May 26, "Muranyi, Arpad" <arpad.muranyi@xxxxxxxxx> wrote:
>
>>=20
>>On the other hand, the reason people turn to encrypted SPICE
>>models is simply because they do not have the time (or don't
>>want to take the time) to figure out how to make behavioral
>>models. If IBIS models were hard enough to make, writing
>>macro models and *-AMS models are even harder. However,
>>SPICE may not be the solution either. For one, it is slow,
>>but I also hear issues with its accuracy (contrary to the
>>popular belief that it is accurate).
>>
>
>Arpad,
>
>What did you hear about the source of SPICE errors? Are the errors due =
>to the failure of=20
>SPICE to accurately handle multi-GHz SPICE device models or something =
>else?
>
>Thanks.
>
>Hassan.
>------------------------------------------------------------------
>To unsubscribe from si-list:
>si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
>
>or to administer your membership from a web page, go to:
>//www.freelists.org/webpage/si-list
>
>For help:
>si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
>
>List FAQ wiki page is located at:
> http://si-list.org/wiki/wiki.pl?Si-List_FAQ
>
>List technical documents are available at:
> http://www.si-list.org
>
>List archives are viewable at:
> //www.freelists.org/archives/si-list
>or at our remote archives:
> http://groups.yahoo.com/group/si-list/messages
>Old (prior to June 6, 2001) list archives are viewable at:
> http://www.qsl.net/wb6tpu
>
>
>
>
------------------------------------------------------------------
To unsubscribe from si-list:
si-list-request@xxxxxxxxxxxxx with 'unsubscribe' in the Subject field
or to administer your membership from a web page, go to:
//www.freelists.org/webpage/si-list
For help:
si-list-request@xxxxxxxxxxxxx with 'help' in the Subject field
List FAQ wiki page is located at:
http://si-list.org/wiki/wiki.pl?Si-List_FAQ
List technical documents are available at:
http://www.si-list.org
List archives are viewable at:
//www.freelists.org/archives/si-list
or at our remote archives:
http://groups.yahoo.com/group/si-list/messages
Old (prior to June 6, 2001) list archives are viewable at:
http://www.qsl.net/wb6tpu
Other related posts: