| Published May
1999 Rubicad News
Philips Microcontroller
Group prefers LADEE over other layout migration tools
An Interview with John Yarborough by Gabriele
Eckert
John Yarborough is Engineering Manager at
the Microcontroller Business Unit of Philips Electronics North America
Corporation in Sunnyvale. He started his engineering career at Navel Research
Laboratory in Washington DC over 40 years ago and then moved to Stanford
Research Institute in Menlo Park, CA. After 17 years in R&D he moved
to the semiconductor industry at Signetics and have been there ever since
working on chip design and managing chip design projects. He is with Philips
for more than 14 years.
???Recently
Rubicad did a layout migration project for Philips Micro Controller operation
on a microprocessor core. Can you tell our readers a little bit about this
project?
We had a microprocessor core that was
laid out in one set of design rules and we wanted to move it to the next
generation process. We needed to combine this core with blocks which
were laid out with the new process rules. We had a lot of effort
invested in the layout of the core to make it as small as possible.
I believe the manual layout took us something like 8 man-years.
Our choices were either to move it to the
new process with a tool like Rubicad’s LADEE, redo it by hand, or do a
new layout with automatic place and route tools. A hand layout would
take another 8 man-years of effort and we estimated that automatic place
and route would double the size of the block. Therefore choosing
Rubicad’s design migration service was the best approach since it allowed
us to save the original effort that we had put into the layout.
???What
are your most important reasons to use a layout compaction method versus
other methods to shift the design to another technology?
In the past when we moved from one process
to another we usually did a linear shrink. But in this case a linear shrink
would be very difficult since we wanted to combine the microprocessor core
with memory which was drawn in the new technology. If you want to
combine an existing design with other circuits in a new technology on one
piece of silicon you run into grid matching problems. In general
the grid winds up being too small to allow masks to be made from a single
database. While it is possible to build masks from two data bases,
one shrunk and one not, it is not easy to make this scheme work without
errors. That was the reason why we chose to use a compaction
tool for the microprocessor core. Rubicad’s LADEE automatically matched
our design to the new design rules on the same grid as the other blocks
in the design.
???What
were the most important factors for you to make this project successful?
There are several very important factors
and it’s hard for me to say which was the most important. One important
factor was accuracy. The design rules had to be met and the design
had to be LVS clean. Also the connections from the core to the outside
world had to remain in the same relative place.
Flexibility was another very important
factor. The LADEE tool from Rubicad let us flexibly scale device
sizes. We needed to scale n and p transistors with different factors, and
in some cases we needed to use a different scaling factor for devices on
a given set of nodes.
Another thing that was important was the
time scale for doing the conversion. It only took Rubicad a few weeks
to complete the conversion.
Last but not least costs are important.
Using Rubicad to do the conversion resulted in a very cost-effective migration.
???What
was the largest benefit for you to get the layout migration done by Rubicad’s
design service?
The largest benefit for us, and this was
based on some previous experience with a different compaction tool, was
that we did not have to learn the tool. Tools, no matter how wonderful,
have a learning curve. Unless you have a lot of migration work to
do it is better to have someone who knows the tool do the job. In
future if we have other migration tasks we will continue to use Rubicad’s
design service rather than trying to do it ourselves.
???Did
the layout area and layout quality meet your and your design teams expectation?
I think they more than met our expectations
in a couple of ways: working with the RUBICAD engineers, we were able to
take advantage of a third layer of metal. We moved from a two metal process
to a three metal process. Rubicad’s engineers moved some of the metal 2,
where we had constrictions on, up to metal 3 in a fairly automatic and
painless way.
If you think of a compaction tool it would
not really deal with new technologies and extra layer of metals but in
a fairly straightforward way the Rubicad engineers did that. In that sense
it met more than our expectations.
It’s hard to say what we expected regarding
the size of the compacted layout. The main thing was that the core
had to fit into the given area of the new circuit. Our expectations
were based on Rubicad’s original estimates. We felt these estimates
were fairly aggressive and Rubicad more than met them.
???Was
there additional manual work necessary to make the layout pass DRC and
LVS which were caused by the layout compactor?
The answer is NO. There is not much else
to say. The layout was 100% DRC clean and passed LVS. We used it as it
existed.
???Can
you tell us what your design team thought about the migration project?
The design team had a hard time in selecting
how we are going to do this project. We had some previous experience
with other compaction tools which was not very pleasant. So we were very
concerned about trying compaction again. When we started talking
to Rubicad they said “No problem. We do this things every day.” What
gave us the most comfort was that Rubicad said they would do the whole
job on a service bureau basis. This made us comfortable since we
didn’t have to learn a new tool. In the end the team was very pleased.
The whole project was done in a week or two, a very short time, and there
were no errors.
???What
are the recommendations you could suggest to other design teams who need
to migrate their design to different technologies or for reuse in silicon
systems?
I am answering this question as someone
who has been involved in silicon design for a long time. The current philosophy
is that you reuse IC design at the HDL level. I think that is not
the best approach if you can avoid it. Even if you have a debugged
high level description you need to go through the complete physical design
process. While it’s better than not having any description, there
is a tremendous amount of work required to go from the high level description
to a physical design. As a result I believe that if you have done
the physical design process once, you might consider saving the physical
design via migration as a means of going from one process to another.
It probably is less painful to migrate than to go through synthesis, place
and route again. If you do a migration most circuits scale without
a problem. For instance clock loading will scale and you won’t have
to go through as many iterations in order to get the buffer sizes adjusted.
I think the odds are much better of getting
working silicon if you migrate your physical design than if you start from
scratch and do new place and route. To look at it another way, if you start
from scratch you need a tremendous amount of effort to get into the same
level of confidence that the new design will work. I have probably shrunk
or migrated more than 50 circuits and more than 90% have worked on first
silicon. I haven’t done that well on the designs I have done from
scratch.
???In
your opinion which meaning will Rubicad’s layout compaction methodology
have in the future for chip design, especially if you think about deep-submicron
technology?
There are two sides to this question.
The first is the one I was making earlier. If you have a part with
a good physical design, take advantage of it. Given the capabilities
of Rubicad’s LADEE tool suite I believe migration is the way to go even
for deep-submicron designs.
The second point is that for deep submicron
designs the RUBICAD tools can be used for more than compaction. For
submicron designs some of its features will really become handy.
There are all sorts of reasons you need to push and massage your layout
in the deep-submicron world. For instance you can spread wires
evenly in routing areas to minimize wire to wire capacitance. Or you can
minimize the loading on specific nets, such as the clock tree. In
the limit you might conceive of going into the layout and optimizing the
size of every transistor in order to make it just big enough for it’s loading.
I believe we haven’t seen the full extent of what Rubicad’s tools can do.
If you only think about a tool to move between processes I think you are
missing a lot.
???If
you have the possibilities to shift a design to next process technology
by new place and route or by layout compaction, which would be your preferred
method?
I prefer to compact it, because the relative
loading stays the same and because it is less work. Today I believe in
layout compaction. In the old days I believed in optical shrinks. I have
done optical shrinks most people would not even think about in order to
save a layout. Today, I believe in compaction for the same reason.
We put a tremendous investment into a layout, even those done with place
and rout tools. As a result if you can recycle a layout, recycle it.
???What
do you think about the idea to use layout compaction technology to solve
timing problems on layout level in a straight way by applying signal specific
rules?
I think it is a much easier and preferable
way than going into the layout and changing things by hand or doing another
place and route. I think we are just going to see what the Rubicad
technology can do for the deep-submicron design.
???How
did you learn about Rubicad’s service and products?
I don’t know. I kept hearing about Rubicad;
I had been hearing about it for so long that I can’t even remember when
I first heard of it. I know it wasn’t through advertising or a show like
the DAC, so it must have been word-of-mouth, I guess I knew someone who
knew about it.
???Have
you done automatic layout compaction before you did this project with Rubicad?
Yes, we certainly did. A couple years
ago we compacted a design, we moved from a process which used EPI to a
non EPI process. This required a large increase in N+ to well spacing.
We used another compaction tool for this project. It was not a good experience.
Since the tool was immature the process took a tremendous amount of work.
Not wanting a repeat of that experience we went to Rubicad this time.
???Which
factors most influenced your decision to work with RUBICAD for your layout
migration project?
The main factor was that we were able
to use Rubicad’s design service on a project base. I talked to others who
used Rubicad’s service here in Philips and their projects were a success.
As a result we decided to go with Rubicad; it was a good decision.
???What
do you like about using Rubicad’s design service?
It’s cost effective, it’s very fast and
the technical staff at Rubicad is very competent. They took our design,
looked at it, and understood it very rapidly. I don’t think Rubicad’s
staff asked as many questions as they told us things about our design and
made suggestions about the best way to do the migration. For example
we were moving from a two metal to a three metal process. Of course
no compaction tool can automatically take advantage of this extra level
of interconnect. We were thinking of moving a certain group of signals
to metal 3. Rubicad recommended moving a different set. As
it turned out that our their approach was better than ours. The most
positive part of my experience on this migration project was working with
the people there.
???What
else would you tell designers regarding design migration?
The bottom line is, if moving an existing
design to a new process think about migration. This is particular true
if you start with a hand-crafted layout, but I think it is even truth if
you start with an automatically place and route layout that came from an
HDL description. I believe the use of compaction tools is not in
the forefront of most engineers’ minds, but it is worth a second thought.
It’s painless, particularly if done on a project basis, and it worked very
well for us
 Back
to Top
|
