( DAC'18 Item 8 ) ------------------------------------------------- [05/22/19]
Subject: Ansys CEO Ajei Gopal cleverly acquires Helic as #8 "Best of 2018"
THE YORGOS GAMBIT: If you take an objective look at the Helic trajectory, it
came as no suprise to me that they got acquired in 2019. The EDA user
interest as shown by word count in the annual surveys easily backs it.
2016: ## 81 words <-- Helic's first Cheesy Must See List mention
2017: ############ 1,153 words
2018: ########################### 2,650 words <--2.3 X
So Ansys announcing they were acquiring Helic in January 2019 made sense.
Buying Yorgos' Helic was a great chess move by Ajei Gopal, CEO of Ansys CEO.
Why? Anirudh just launched Clarity; which does board, package, and silicon.
While Ajei's HFSS is for board and package -- it simply CAN NOT do silicon.
Helic does silicon -- plus it has a proven track record for speed & capacity
in silicon -- which means Ansys HFSS *plus* Helic will (after some massive
integration) make for interesting speed chess between Cadence and Ansys.
Here's the initial game step-up with the first few moves...
Ansys (HFSS) Cadence (Clarity)
---------------- -------------------
market position 3-D EM FEM leader newbie under 1%;
2,800 customers 2 or 4 beta
worldwide users worldwide
strengths dominant market claims cloud, multi-
leader with threading, 12x to 14x
entrenched user less memory, scales,
base 7x to 12x faster
features HFSS board and Clarity claims board,
package only, no package and silicon
silicon -- but w/
Helic does silicon
weakness HFSS is old code; Clarity is unproven.
memory hog, no lots of claims.
cloud, single
machine use only
remedy in 24 to 36 months in 24 to 36 months
HFSS has be cloud, Clarity must publicaly
multi-everything prove itself
weakness it won't be easy it won't be easy
integrating HFSS getting Clarity out
with Helic of beta
Or said another way, because Ajei acquired Helic, it's now speed chess on
which R&D team (CDNS or ANSS) can code faster and better to fix their own
specific tool problems. Prior to Helic, Ajei's HFSS didn't have silicon.
But now with Helic, it's anyone's game. (Clever man that Ajei!)
"Do. Or do not. There is no try."
- Yoda, mystical Star Wars muppet (1980 - )
"It is not the strongest of the species that survives, nor
the most intelligent that survives. It is the one that
is most adaptable to change."
- Charles Darwin, English naturalist (1809 - 1882)
HELIC ALSO HELPS REDHAWK: On a side note, Apache Redhawk has always done RC
extraction internally. With Helic, RedHawk will add induction -- making it
an internal RLC extraction -- thus more closer to real life 7/5/3nm physics!
---- ---- ---- ---- ---- ---- ----
---- ---- ---- ---- ---- ---- ----
---- ---- ---- ---- ---- ---- ----
QUESTION ASKED:
Q: "What were the 3 or 4 most INTERESTING specific EDA tools
you've seen this year? WHY did they interest you?"
---- ---- ---- ---- ---- ---- ----
VeloceRF -- Inductive Device Synthesis & Modeling tool
We nominate Helic VeloceRF. It's user-friendly and quickly
generates the physical layers and electromagnetic modeling for
a wide range of our RF structures.
We think it's best in class at combining accuracy in the models
and efficiency for our RF design methodology.
- Fast and accurate extraction times
- Optimization engine generates structures based on a range
of the user's design constraints
- Parameterized Pcell-like structures that can be placed
as an instance in a layout and changed on the fly
- It can place multiple structures in a layout and
extract a comprehensive model
- Physical layers are automatically DRC clean, including a
selectable tiling map for density requirements
- Generation of RFM format models for simulators. This
is a growing trend in the industry (rather than SP models)
and Helic is ahead of that trend.
I'd highly recommend VeloceRF for RF or high-speed design groups that
need good tools that work.
---- ---- ---- ---- ---- ---- ----
Helic RaptorX
We use Helic RaptorX electromagnetic modeling tool. RaptorX can
extract and create a model for various layouts like clock routing,
dense on chip signal routing, complex multi-port spiral Inductors,
and top-level power grids.
We use RaptorX for RLCK or full EM extracted models of all the above.
It provides netlist-able schematic and symbol view and n-port models.
We use Helic's RLCk netlist models for transient analyses.
General feedback:
- Performance. Helic's complex structure extraction is very
fast. With proper memory resources, the extraction time can
be 20% of the time as comparable EM extraction by other tools
-- or even faster.
- Accuracy. We extract complex multiport high-speed clock
routing, signal routing with proper aggressors and spiral
inductors. We believe Helic's accuracy range is believed to
be within 1%-2%, based on deduced silicon correlation.
What we liked:
- Point-and-click based net selection process. You can select a
net and choose the type of extraction, frequency requirements,
etc. from the RaptorX GUI.
- Layout labels work as port definition. Anything not connected
and not labelled is counted as floating and discarded by the
tool during extraction.
- For EM analysis, you provide frequency for extraction, rather
than defining boundary conditions or special types of ports.
- RaptorX is well integrated with Cadence Virtuoso ADE. We've
had no issues.
I recommend it for extraction of complex and critical routing,
complex power grid analysis for evaluating switching noise etc. and
for complex spiral extraction with multiport systems.
---- ---- ---- ---- ---- ---- ----
Helic VeloceRF
VeloceRF is a powerful transformer and inductor synthesis tool. You
give it inductance, the Q, the SRF, the size, input/output impedance
as inputs.
You can then use it to generate DRC/DFM clean inductors and
transformers, as well as the corresponding models that can be used for
transistor level electrical simulations. You can easily modify and
optimize your devices in terms of any parameter.
We have confirmed designs up to 60GHz against other 3rd party EM tools.
VeloceRF provides you with a complete model of all the inductors and
transformers of your circuit, including coupling. The model can be used
during electrical, pre-layout, and simulations. So, you can optimize
your floor planning and minimize your crosstalk before the detailed
layout.
You can also use it to design different kinds of transmission lines
after selecting Zo, top and bottom layers, etc. and evaluating the
performance.
We absolutely recommend it. We have used VeloceRF together with Cadence
Virtuoso, Mentor's Calibre and TSMC 65nm MSRF kit. The integrations are
very easy and work perfectly.
---- ---- ---- ---- ---- ---- ----
RaptorX - Pre-LVS Electromagnetic Modeling
It accurately model custom layouts that have arbitrary geometries
(i.e. it is not limited to spiral-like structures).
This lets you create and experiment with your own unique ideas for RF
structures as well as properly modeling the effects of transmission
lines or the coupling effects of nearby signal aggressors.
The LVS module feature is seamless for creating the necessary Cadence
cellviews, so a structure can permanently be placed into the schematic
and layout without affecting the LVS and postlayout extraction flow.
(With other tools, our design teams must sometimes go to great
lengths to massage their flow and manage multiple versions of layouts
and schematics when including RF structures.)
We like that RaptorX can take a structure generated with VeloceRF,
automatically flatten the layout to make it custom, and then
generate an EM model.
For example, a T-coil is optimized and generated with the VeloceRF.
This structure could then be converted to the RaptorX flow and then
modified to include its metal leads to the rest of the layout or
the impact of an overlapping or nearby bump/pad.
I highly recommend RaptorX.
---- ---- ---- ---- ---- ---- ----
Helic RaptorX
We plan to start using it soon.
---- ---- ---- ---- ---- ---- ----
Helic RaptorX is an EM simulation tool for extracting layouts, e.g.
inductors and routing parasitics, and is well integrated in Cadence
Virtuoso.
In general, RaptorX is very user friendly and easy to set up, e.g.
simple port definitions with point and click placement and no special
settings regarding port types / boundary conditions. Even large layouts
simulate quite fast and simulation scales well with # cores, in addition
the simulation method/meshing can be optimized on layer/net fine basis.
Our circuit performance measurements indicate that its simulated
inductances match well.
---- ---- ---- ---- ---- ---- ----
I use both Helic VeloceRF and RaptorX. Here is my input.
1. Helic is very user friendly. It is easy to use, even for
people with little EM background;
2. Simulation is pretty fast;
3. It can handle large size EM cases.
---- ---- ---- ---- ---- ---- ----
VeloceRF does EM simulation of Helic p-cells. P-cells can be
parametrized in the Cadence layout view and directly simulated
(L and Q).
It enables easy sizing of inductors and optimization of parameters for
best Q. It gives you several parameters to optimize the layout for your
specific application case.
Highlights:
- Easy optimization of inductor parameters for best Q
- Different device geometries available as p-cells to support a
variety of use cases
- Good Cadence integration
The tool is very useful for evaluation of different inductor topologies
in an early design phase and optimization of the final design regarding
different parameters.
---- ---- ---- ---- ---- ---- ----
The Helic tools, VeloceRF and RaptorX, are easy to use. In our
experience the layout of the circuit provides the geometry to the
simulator while the material parameters, needed for all simulations,
are taken from the backend file so the user doesn't have to worry
about it. This means that the tools are used in supported PDK's.
1. VeloceRF.
VeloceRF synthesizes inductor and transformer structures.
After design goals are specified, a series of possible
design options are generated in a short period of time.
The resulting design options are all DRC clean and
tiling can be included.
Generally, after we have a design, we will simulate it,
including tiling, with RaptorX. It seems that the layouts
may have some restrictions due to limits in the cross-over
geometry. This, in turn, limits the geometries.
2. RaptorX.
RaptorX is a powerful analysis tool that is easy to set up and
use. In my experience it is easier to use than most EM tools.
The resulting outputs include s-parameter files and a lumped
element netlist. While the s-parameter files can be used for
steady-state simulations, the netlist permits time dependent
simulations.
The general nature of RaptorX makes it useful in the general
analysis of metallic structures including inductors,
transformers, capacitors, transmission lines -- and coupling
between structures.
While we have observed that the simulation time of the Helic tools can
be quite 'fast', this really depends on how complicated the problem is
to begin with. Simple isolated structures can be simulated in minutes.
In contrast, complete test circuits including pads, ground planes and
tiling can easily take hours to simulate. However, it must be pointed
out that setting up the problem for simulation is relatively simple.
While in the past we have been successful with these tools at
frequencies below mmW, we are still evaluating the tools for mmW
applications. We've used RaptorX to understand the results of
certain problems. For example:
- We used it at mmW frequencies to show that the close proximity
of a ground plane caused significant degradation of an
inductor's performance.
- It also demonstrated where there were questionable test
results because the data was shown to not follow expected
physical behavior.
Support has been very good. The Helic staff have worked closely with us
to analyze some complex problems. This has included interpreting
results and setting up various problems for most efficient simulations.
---- ---- ---- ---- ---- ---- ----
Exalto Post-LVS RLCk Extraction
We tried Helic Exalto to capture nets, which were not part of momentum
sim.
Performance was good, and very close to other EM tools in all terms:
speed, accuracy, etc.
We don't currently use it for sign-off; however, we may do so in the
future.
Exalto has good integration with Virtuoso, you always know that the
simulated layout is a tapeout version
---- ---- ---- ---- ---- ---- ----
Helic VeloceRF is an inductive device synthesis/modeling tool that
generate inductors and transformers for RF/Microwave design.
- We've found the accuracy (e.g. to millimeter-wave frequencies)
is sufficient for a 1st cut design.
- Synthesize devices with tight physical constraints -- this is
good for inductors.
- It is well-integrated in Cadence Virtuoso ADE.
I recommend it for inductor design -- VeloceRF is very fast to get the
desired inductor. I'm looking for improvements for transformer design
to make it easier to get the desired result.
---- ---- ---- ---- ---- ---- ----
We use Helic VeloceRF for inductor and capacitor analysis for custom
layout.
Its performance is good. The accuracy for mm-wave frequencies is not
always good; although we did not perform a systematic comparison, for
some structures there were differences (inductance, Q-factor). We did
not debug the root cause of the difference.
VeloceRF has a good integration to Cadence Virtuoso. Helic can improve
its integration to ADE-L/Maestro.
---- ---- ---- ---- ---- ---- ----
We are introducing Helic RaptorX into our RF LDMOS design flow. There
are two features of this process that make it different from a typical
RF CMOS process.
First, we make RF power amplifiers in it, with the output power of a few
hundred watts from a single chip. For this we need really big
integrated components. That has a big impact on the layout parasitic
extraction and possibly makes our perspective quite different than that
of a RF CMOS designer.
Second, the RF LDMOS process is built on very low ohmic substrate. Our
understanding is this is not an application area for which RaptorX has
been intensively tested.
We should be always cautious when comparing run times of simulation
tools. For each tool we choose the speed/accuracy trade-off
differently. Comparing the runtimes is usually easy -- it is just one
number. Comparing the accuracy can be tricky. Discussion on the
absolute accuracy quickly becomes academic since what we can only
reliably measure on silicon small parts of the circuitry, too small to
tell much about the performance of the whole product.
So, to make it more practical, we want the inaccuracies induced by the
extraction tool to be significantly lower that the design margin
(distance between the simulated performance and the requirement spec).
We are not yet done with this part, but I can share the comparison on
the component level and at the block level.
We have verified the accuracy of RaptorX against RF measurement using a
range of components like inductors and capacitors. At this level the
accuracy was very good. RaptorX can mix-and-match different accuracy
levels in one run so that you can choose them individually for each
component. The cost tradeoff is the manual work you need to do prior
to the simulation, such as partitioning the layout and setting up the
simulator.
The simulation time of inductors was very much comparable with the
simulation time of our method-of-moments simulator. However, we noted a
big difference in the runtimes for large MIM capacitors and even bigger
for fringe capacitors. These components were simply not simulate-able
with the method of moments.
So, our speed improvements range from no improvement to infinitely large
improvement. We are looking forward to further improvements in the
network order reduction. Currently this phase of extraction is the most
significant contributor to the extraction time and requires a huge
amount of RAM.
When it comes to the extraction at the block level we observe 4X to 6X
shorter times than the method of moments with only a limited impact on
the accuracy. This makes a big difference to us. The nice part of
this is that we can choose even much faster extraction, like 30 times,
if we decide to give in on the accuracy. The absolute accuracy remains
a concern. It is a complicated question since all we can measure is
the performance of the entire product. This includes the active
components and the packaging, so it is not easy to draw a definite
conclusion about the error budget.
Almost all the extraction modules can run on multiple CPU's. In the
coming weeks we are going to study the benefits of using even more
powerful hardware.
RaptorX copes very efficiently with large metal shapes full of slits.
Also stacked vias are dealt with very efficiently. We communicated to
Helic that we connections to substrate play significant role in silicon
processes based on low-ohmic substrate. We hope that this feature will
be implemented soon.
The RaptorX UI is easy in use but we miss a few features like storing
the net-level extraction settings for re-use in future extractions. Our
long-term target is to integrate RaptorX within ADS. The port and net
properties should be set from the ADS UI.
RaptorX doesn't require you to define boundary conditions or special
types of ports. This simplicity can be an advantage or a disadvantage.
Setting up a simulation is indeed very fast. However, we see however
some limitations of pin ports. Our integrated components can be so big
that it is no longer possible to consider the ports as 0-dimensional
objects. The whole 200um edge can be a pin. It is something to talk
about with Helic.
We see a gap between RaptorX and the method of moments we use now.
RaptorX is not able to show current density over the layout.
Connectivity check is possible but cumbersome. The mesh editor shows
pins (sometimes hard to find!) but not the pin names. It is not
possible to make cross section through the layout.
There are a couple of other Raptor features that we haven't used yet,
but that look useful, such as the what-if analysis to evaluate of
crosstalk among blocks by addition or exclusion of nets. Also, it
supports batch runs for routine jobs -- we expect to use that a lot.
---- ---- ---- ---- ---- ---- ----
Related Articles
SCOOP! -- Anirudh goes total war on Ansys mothership at CDNlive'19!
Helic Exalto gets #7 Best of 2017; because inductance counts at 7nm
Helic's accidental pre-DAC marketing attack on Anirudh's Pegasus
CLKDA Isadore warns "Don't drink the SNPS/CDNS timing Kool-Aid!"
Join
Index
Next->Item
|
|
