( DAC 01 Item 34 ) --------------------------------------------- [ 7/31/01 ]
Subject: Simplex & Simplex 'X Technology'
YESTERDAY'S TECHNOLOGY TODAY: Trying to build upon the success of its
backend analysis tools, Simplex made their Year Of 45 Degree Router. This
has definitely wowed the newbies, but many of the backend veterans have
voiced some serious concerns about it being much harder than it appears
at first. Apparently 45 degree routing is an old idea. The reason why
Simplex is offering it as a service and as a "technology shared with
partners" (instead of just incorporating 45 degrees in a commercially
available router) *is* because it's so difficult to use.
"We are using Fire & Ice from Simplex and have been happy with it."
- [ An Anon Engineer ]
"We used Simplex VoltageStorm on our last ASIC. Very detailed analysis,
but only as good as the power estimation that you put in. The big
question is how to come up with better power estimates?"
- John Busco of Brocade
"I've had an enormous amount of agony getting Simplex VoltageStorm to
run. It seems to be peculiar to something I'm individually doing;
other guys in my group have had no trouble. But their support is
absolutely first-rate: very responsive, some of the best I've seen
from an EDA company. So although I'm ready to put my fist through
my monitor, I'm actually not annoyed with Simplex itself, and in fact
I'd recommend them generally."
- Natalie Overstreet Ramsey, Vitesse Semiconductor
"The first user interface several years ago for Simplex was horrible,
basically just a bunch of standalone scripts and a huge sloppy manual.
I never looked at them again to see how they improved it. I'd rather
use a slightly less accurate tool with an easy user interface than a
hard-to-use tool that has a better core algorithm. Time-to-market
is more important than absolute performance."
- [ An Anon Engineer ]
"We use Simplex and Moscape. Looking into Primetime-SI."
- [ An Anon Engineer ]
"Simplex? We like and it find signficant numbers of SI problems with
it. You will struggle if you do anything strange (EE, SONOS, BiCmos).
MachTA is still a ways off. Maybe 1 more DAC & it will be there."
- [ An Anon Engineer ]
"I drooled over the Simplex 'X Architecture' announcement, and the
related Liquid Routing technology. Simplex has done something
'basic' but utterly compelling in terms of net results. It's times
like this that I regret my policy of not buying shares in companies
that I do business with! In my view, Simplex has a major winner on
their hands. Simplex is offering an opening to other EDA companies
to participate through the X Initiative consortium. But whether
companies like Synopsys can tolerate such a blow to their corporate
'pride' is an open question. If not it'll be a pity, because I
don't think that Route Compiler is a winner as it currently stands."
- Mike Carter of Mosaid Technologies
"Simplex's X architecture announcement sure got my attention. I and a
collegue had looked at this briefly 5 or 6 years ago (the concept has
been around for much longer). At the time, we concluded that the
effects of doing diagonal routing were much more far-reaching than
just the router -- wireloads (physical synthesis didn't exist as a
product at the time), placement, global route, clocking, and extraction
would all be affected. Not to mention the issue with losing 30%
density in the diagonal layers so that you'd line up with vias on the
horizontal and vertical layers. Simplex claims to have solved the
problems, and supposedly has the code to implement the solutions.
Their 'gridless octilinear' routing claims great things - I'm very
interested to see if it gains acceptance.
- Mike Berry, Silicon Logic Engineering, Inc.
"Like the Simplex 'X' technology with Toshiba. Hoped the other vendors
will buy in as well. Interested to see if there is a new extraction
algorithm for this 45 degree router, and how they can address the
problem with scanner mask generation.
- [ An Anon Engineer ]
"Simplex X Technology has not shipped yet, and will not be decently
usable until incorporated into somebody else's suite. What is
their plan?"
- [ An Anon Engineer ]
"Future: 1. X-routing: using 45-degree routing
- used in PCB routing for years, but not in chip routing because of
limitations in the manufacturing process. These limitations
apparently don't exist today.
- has big potential benefit in terms of decreasing wire length, and
thus improving speed, area, power, etc.
- but requires a big change to the flow: PDEF, for example, only
supports Manhattan-style routing; affects parasitic extraction,
floorplanning (power & clock routing), mask generation, etc., etc."
- Kris Monsen of Mobilygen Corp.
"Simplex:
I found more substance in Simplex' X-architecture initiative than I
expected. Most of their work to support the initiative can be
classified into two parts. One is working with mask-generation and
manufacturing to support the initiative. The other is new routing
algorithms. Their routing directions are HVHDD'. However, when I
examined a routed design, I saw a lot of small, compact regions where
the routing is free-form, like PCBs or river routing. The diagonal
routes allow them to get around worst-case scenarios for river routing.
Such routes also minimize number of vias. They also use octagonal
vias, instead of square, for Manhattan-diagonal intersections. I also
discovered that two key developers from the University of Bonn CAD
group, responsible for routing and timing, joined Simplex over the
past 12 months. The Bonn tools were used by IBM-Boeblingen to design
mainframe processors using an ASIC-like methodology.
The tools gave very good results, but needed hand-holding. In a
services-oriented business model, this should not be a big drawback."
- Deepak Sherlekar of In-Chip Systems
"Non-orthogonal routing capability has existed in Cadence's IC Craftsman
since it's inception back in 1995. I don't think that Simplex fully
appreciates the magnitude of the challenges that they are facing. I
believe that it's going to be a lot harder, and will take a lot more
time, than Simplex realizes to get this technology into even the most
advanced designs. Even if they can route a DRC/LVS correct design
using 45 degree wires, can they extract it, verify its functionality
and timing, then manufacture it with acceptable yields?"
- [ An Anon Engineer ]
"Simplex's X-initiative is only a concept right now. They have built a
prototype router which they can use to do some design service for their
special customers. Even if it evolves into a real general purpose
router, it is still unclear how the whole flow will come together.
They have to get some serious partnerships (which is hard) or buy a
bunch of companies."
- [ An Anon Engineer ]
"Simplex X seems to me a lot of hype over something that has been known
about for a long time. Would be great for the EDA industry if it takes
off as they could charge for all new layout and verification tools in
this area. I think the algorithms involved in today's layout and
related design rules, parasitics, etc. are already too complex for
the EDA companies to handle/estimate and calculate accurately. I hate
to think of how many new problems are caused by the geometries involved
in X layout. Having a diagonal distance between two points is, of
course, shorter but in a congested chip will the layout algorithms
actually order things that well to take advantage of it?"
- Scott Evans of Sonics, Inc.
"So far we're sticking with Apollo. With PhysOpt and the NEW batch
of physical synthesis tools, routers may not be king of the hill in
physical design any more. A dumb maze router, or even a spiffy
45 degree router may not mean much to guys that want to close timing,
generate clocks and scan, and then meet power, electromigration, IR
drop and signal integrity rules in less time than they just do routing
now.
Simplex 'X Technology' sounds like 'back to the future' to me. There
were some 45 degree tools a long time ago. They never quite worked
and when they did they never gave anywhere close to the possible
improvements in density that their developers promised. Is this time
different for some reason? Will this stuff work with placement tools
that expect the router to pick a Manhattan shortest path?"
- John Szetela of AMD
"I can talk about the Simplex stuff. The idea is very old, but can't be
used until you have enough layers of metal for routing. You basically
need at least two layers to minimize the logic cells below. The next
two are naturally horizontal and vertical routing layers. The thick
top metal layer(s) are for clocks and power. This is already at least
5 layers, and some people like to use one more for the logic below, and
one more thick layer for more power and clocks (7 total). To make use
of diagonal layers, you need at least 4 layers devoted just to routing.
This is really only an option in the new 7 and 8 layer metal processes
coming on line now. In the past, I worked a bit with the idea of three
routing layers at three angles (0, 120, 240 degrees). It didn't work
out very well, since there was no way to go up and down efficiently.
However, with 4 layers, that problem is solved.
I see a big problem for standard cell and gate array routers being that
the global router doesn't know how wide the horizontal channels will be
(it's determined by the detail router). This messes up the diagonal
routing grid. All of my routers are fixed width channel routers, so
they won't have this problem.
Given that I have 4 layers for routing, I would use 2 of them diagonal
layers (45 degrees to the left and right). The impact on the routing
tools I write would be minimal. I certainly hope the Simplex guys
don't think they invented diagonal routing, because they didn't. I
plan to do it when it makes sense for the products I support."
- Bill Cox of VI ASIC
"This "X Technology" could be the next Physical Synthesis or it could
be the next Behavioural Compiler. It remains to be seen."
- Andrew MacCormack, Tality/Cadence
|
|
