( ESNUG 529 Item 5 ) -------------------------------------------- [07/18/13]
Subject: Invarian CEO insulted after being compared to dying Apache/Ansys
> Invarian Pioneer Analysis are wannabes who compete head on against
> Ansys/Apache with a mix of very-longly-named-fragmented-tools that
> collectively do power, EM, IR-drop, thermal for chips/package/PCB.
> Claims 40 nm and 28 nm "within 2% of real physical measurements."
> (booth 1332) Ask for Vladimir Schellbach. Freebie: rulers
>
> - from http://www.deepchip.com/gadfly/gad053013.html
From: [ Jens Andersen of Invarian ]
Hi, John,
1. We don't do PCB's. Never did. That's simply factually wrong.
2. How dare you call us Apache wannabes!
At one time 12 years ago Apache had a "hot" technology (for it's day), but
it's core data structures and key algorithms all originate from late 1990's
computer science. Because of limited CPU and memory at the time, Apache's
fundamental SW architecture meant their EDA tools had to solve power, EM,
IR-drop, thermal in a PIECEMEAL way. They gave the customer a new tool each
year for each separate part of the puzzle (RedHawk, PowerGate, FAO, NSPICE,
PsiWinder, Sentinel, Totem, PathFinder) one at a time.
PIECEMEAL VS. SIMULTANEOUS
But an awful lot has happened in computer science since the late 1990's.
Now there's many more CPU's, much more memory, and tons of raw bandwidth
available -- which opens the door to new data structures and new algorithms
that let us create EDA tools which SIMULTANEOUSLY solve power, EM, IR-drop,
thermal, and transients concurrently -- and with less slop than what an
Apache piecemeal approach does.
SIMULTANEOUS is what Invarian Pioneer Analysis does.
Unfortunately, Apache's "piecemeal" approach is the status quo today, with
it's set of well-developed, patched, and aging tools that are currently
coming to the end of their life cycle. At or below 20 nm is where the
trouble begins. Apache cannot keep up. Their aging tools were never meant
for handling this much physical complexity. Adding to the complexity is
3D of both transistors and stacked-die (micro and macro); thermal issues;
leakage currents that are dependent on temperature and tighter geometries;
and increasing current densities.
Because Invarian is a new SW architecture using 2012 technologies, our
Pioneer tools can go sub-20 nm and the designer gets a clear view of what
is happening -- down to the sub-transistor level -- so intelligent trade-
offs can be made before its too late on performance, power and price.
We do comprehensive hierarchical, run on multi-core machines, and we scale
linearly to 16, 32, 64, 128 cores, etc. Apache's trying to catch up.
An Apache user pays a hefty price of over-design and unpredictable tape-out
delays due to painful non-converging repair loops. Our tool is physically
correct-by-design so designers see real power, EM/IR and thermal problems
as they exist before tapeout.
PAST OR FUTURE?
To make matters worse, Apache has been mostly treading water ever since it
got acquired by Ansys three years ago. Ask yourself when was the last big
new tool development that Apache has announced? It was in 2010 -- right
before their Ansys acquisition -- with their PathFinder ESD tool. Since
then it's been hanging on by inertia.
Let's say you have an 8-billion transistor design. How do you simulate it
in a reasonable period of time? Invarian's new macro modeling solves this
capacity issue while keeping runtimes short. We do not sacrifice accuracy
in order to achieve short runtimes.
Apache fails miserably as designs grow in size.
Currently our customers have physically measured results back in 40 nm,
28 nm and 20 nm and we are within 2% of real physical measurements.
We have customers who tell us that their designs from Pioneer correlated
with a few percent of actual silicon -- and were more accurate than their
SPICE results! Can Apache say that?
EASE OF USE
Simply put, Invarian Pioneer tools were crafted from day one to be easy
to use. Older, piecemeal products like Apache tools only now give great
results when the Apache AEs are running them -- especially on the newer
and more complex process nodes. And if you don't get their "star" AE in
your Apache tool contract, you run the real risk of having your designs
over-margined -- which costs money!
In contrast, Invarian Pioneer Analysis doesn't need AE's. It's new and
was designed from the beginning to be easy to use. There is NO "star" AE
dependency with us. You do not have to be a Jedi Mind Master to use
Pioneer. It just works.
INVARIAN IS NOT APACHE
Apache today is falling into the end-of-life product cycle. So when you
say we're an Apache wannabe, John, we take offense. We're the better next
generation EDA tools taking our rightful place at the sub-20 nm table.
- Jens Andersen
Invarian, Inc. Sunnyvale, CA
Join
Index
Next->Item
|
|
