( DAC'19 Item 6a ) ------------------------------------------------ [04/17/20]
Subject: Three MENT Solido vs CDNS ADE-XL user benchmarks is Best of 2019 #6a
SOLIDO'S UPSIDE SURPRISE: Solido first popped up on my radar nearly 10 years
ago, in 2010. (See DAC'10 #2) Now, 10 years later, users are barking about
Solido's Variation Designer using "machine learning" sampling to reduce the
number of SPICE runs needed to N-sigma circuit analysis. Over the years
Solido always had loyal happy users, leading up to it being acquired by
Mentor in late 2017.
Even so, I was caught off guard in my Best of 2018 report when Solido-Mentor
"beat the odds" of a hot EDA company fading into the woodwork after it got
bought out.
Frequently post-acquisition stories are notorious shitshows -- but some weird
voodoo that Joe Sawicki, Ravi Subramanian, Amit Gupta, and Tony Hemmelgarn
managed to work out behind the scenes made Solido explosively grow with the
customer base both inside of Siemens itself and outside of Siemens with the
external EDA user base. In fact, Solido earned my #1 spot for 2018,
with 2,748 words from happy users -- more than double the prior year.
- from http://www.deepchip.com/items/dac18-01.html
AND THE BEAT GOES ON: Solido's technology and support continues to thrive in
the Mentor-Siemens captivity. In this year's Best of 2019 report, even more
happy users wrote in with nearly 4,500 words.
2019: ############################################# 4,490 user words
2018: ########################### 2,748
2017: ############# 1,282
SOLIDO TUG-OF-WAR WITH CADENCE ADE-XL: on the tech side the users were doing
the usual gushing about doing 249-sigma analysis with Variation Designer,
but what really caught my eye ...
... were the 3 hands-on user benchmarks comparing Solido vs. Cadence ADE-XL
(Assembler) -- and it looks like Tom Beckley has some competition nipping
at his Virtuoso ADE-XL monopoly.
"Since Solido covers all the use cases I need, I primarily use it.
There is no reason to use Cadence ADE-XL, as it gives comparable
or inferior results."
"Speed -- Solido is 75% faster than ADE-XL."
"Solido's Monte Carlo tool is much better than Cadence ADE-XL,
and the data representation methods Solido uses are much easier
to understand."
---- ---- ---- ---- ---- ---- ----
---- ---- ---- ---- ---- ---- ----
---- ---- ---- ---- ---- ---- ----
QUESTION ASKED:
Q: "What were the 3 or 4 most INTERESTING specific EDA tools
you've seen this year? WHY did they interest you?"
---- ---- ---- ---- ---- ---- ----
USERS BENCHMARK SOLIDO VS. CADENCE ADE-XL (ASSEMBLER)
Solido Variation Designer
Solido Variation Designer lets us evaluate our design functionality
from a process variation point of view. It accounts for the foundry
process parameters, so we can evaluate our design performance quickly
across voltage and temperatures.
We do IoT analog blocks and need design reliability within 4-sigma.
Mentor-Solido Variation Designer vs. Cadence ADE-XL (Assembler)
1. Speed -- Solido is 75% faster than ADE-XL (Assembler)
2. Variation Designer is more straightforward to use. The way
Solido arranges the plots, the data, and the reports makes
more sense and is more intuitive.
3. Variation Designer has more detailed insights/features vs ADE-XL
The most important aspect is the trend distribution analysis. Usually
process variation follows a Gaussian shaped distribution. However, in
some cases it doesn't follow that bell shape, and is harder to debug.
Solido gives you a way to see that trend, while Cadence ADE-XL assumes
a Normal distribution, which can lead you in the wrong direction for
troubleshooting.
I like Solido's Variation Designer and recommend it to my peers.
---- ---- ---- ---- ---- ---- ----
We use Solido Variation Designer Fast Monte Carlo and PVTMC Verifier.
Both take in your SPICE netlist; their output are distributions or data
tables with simulation results, that take into account the process
variations based on the spice model parameters.
Solido's PVTMC Verifier can run 4.0-4.5 sigma variation analysis with
a seriously reduced number of simulations.
We tried a regular Monte Carlo approach, but it took too long to
simulate to 4.0-4.5 sigma. So, we would run a few 100 simulations
and extrapolate. Our results seemed close, but we usually couldn't
confirm them, as it would take too long.
Solido's claim is they can simulate the data points without
extrapolation. They have a sampling algorithm to choose the best
samples out in the 4.0 to 4.5-sigma distribution rather than just doing
random simulations -- so, we'd only need 1,000-2,000 simulation runs
for our 4.0 to 4.5 sigma analysis.
Our methodology is:
- Use Solido's PVTMC Verifier to verify our design to 4.0-4.5 sigma
across a number of corners
- Then run Solido's Fast Monte Carlo at a few corners to confirm.
We like this redundancy.
OUR COMPARISION OF SOLIDO VS. CADENCE ADE-XL (ASSEMBLER)
Visual Reports
Solido's visuals are useful -- we use them a lot for documentation and
analysis.
- The NQ plot (normal quantile plot), should look like straight
line if you get the expected Gaussian distribution. If not,
it will bend up or down.
- We can see this trend without completing our simulation.
GUI's
Solido's GUI looks good and has been improving. Compared to ADE-XL,
the Solido one has an advantage.
- Solido plots the data point on the histogram as the data is
being processed, so our team can look at partial data.
- We can also export those partial data points to CSV and Excel.
- This is important because with a few 100 runs, we may have IT
issues to prevent further simulations and need to show data in
a short time.
- Solido can do this because they use a post-processor to extract
data points. Thus, the data is a text format, so simpler to
recover.
- Solido's display interface does occasionally crash, but Solido's
support team is pretty responsive in acknowledging it. Fixing
it could take time as they must be able to replicate it.
Even so, because our simulations still keep running in the
background, I can just close the interface and look at the partial
results. (We only lose the rest of the intermediate observation.)
- In Cadence ADE-XL if it crashes, it can't show any data. We
lose that partial data that is important for team discussions.
- For example, if we run a few 100 simulations over the weekend
and it crashes, ADE-XL randomly decides when to make it usable.
- Cadence has this limitation because its Parameter Storage
Format (PSF) tends to corrupt itself if the tool crashes
unexpectedly, so ADE-XL can no longer extract data.
- Cadence is a bit stuck here because they must maintain
compatibility with the many tool generations they have
released.
Compute Resources
Solido also has an advantage as to the way their tool runs.
- Solido will auto-delete your netlists or work files that
take up space but don't add value. This lets us more easily
run 1000s of sims with our limited compute resources.
- Cadence ADE-XL stores everything, so we run out of disk space
fast, especially with post-layout netlists that are >100M.
ADE Integration
There is one thing that Cadence does better than Solido; it can more
easily make a family of plots to see how the waveforms vary across MC
iterations (e.g. plotting the voltage vs. time). For example, we
might use this visual representation to show designers the limit on a
particular voltage or current of interest.
Solido does NOT do as well there because of how Cadence deals with
3rd party tools in their environment.
- When Cadence changes their waveform viewer, the old hooks
break.
- Plus, for example, I can use Solido on ADE, but it's not
permissible ADE-XL.
- CDNS has incentive to make it harder for all other vendors,
i.e. it is not just Solido.
However, I can see signs of Mentor now trying to push their own platform
with AFS simulator, i.e. working on ADE-XL alternative. That would
give Solido more even footing to compete on integration also.
Analysis Features
- Solido's PVTMC Verifier is unique; there is no equivalent in
ADE-XL. It saves us a lot of simulations when analyzing our
corners.
- For normal Monte Carlo analysis, I tend to use Solido more
often than ADE-XL because Solido has more options -- e.g.
PVTMC Verifier, and basic MC.
Since Solido covers all the use cases I need, I primarily use it.
There is no reason to use the ADE-XL, as it gives comparable or
inferior results.
---- ---- ---- ---- ---- ---- ----
I mainly use Solido for functional verification and Monte Carlo
simulations to ensure design robustness.
Solido's Monte Carlo tool is much better than Cadence ADE-XL, and the
data representation methods Solido uses are much easier to understand.
It is nearly impossible to achieve the coverage in ADE-XL that we can
get with Solido if we only used our AFS and Spectre SPICE simulators
alone.
- Normal full coverage Solido simulation will take about 3-4
hours in my test case.
- If we used a normal Monte Carlo tool (ADE-XL) it would take
days to finish.
For prior simulations, our design requirement accuracy was normally
3-sigma PVTMC.
However, after we adopted Solido, our design coverage has grown up to
5-sigma. It can catch every corner case that happened in our silicon.
For example, it replicated a 5-sigma circuit weakness in our silicon
and helped us to locate the bug and fix the circuit for production.
Solido's GUI is ok, but it is on the slow side -- it needs to be more
responsive.
My overall impression is good. Solido is the best Monte Carlo tool
that I have used, and I'd definitely recommend it.
---- ---- ---- ---- ---- ---- ----
---- ---- ---- ---- ---- ---- ----
---- ---- ---- ---- ---- ---- ----
USERS ON JUST SOLIDO ALONE
I'm using the HSMC module of (Mentor) Solido Variation Designer.
- It is a high sigma Monte Carlo analysis tool and helps us to
understand our designs' 5-sigma and 6-sigma quantile values
well without having to run 100,000 SPICE simulations.
- HSMC's inputs are the same as for any SPICE simulation, and
statistical information output is convenient for post
processing.
We use it for 5-7 sigma analysis.
- The run time is fast, my simulation finished in 30 minutes
instead of 2 days.
- The tool's accuracy is reliable. We run SPICE simulation with
a high number of the samples to compare the data.
- Their GUI is good -- you can visualize the data distribution
and debug if the simulation did not go through.
- My simulation does not have that many variables, but I never
see any scalability issues.
I have been using Solido for more than 7 years. It generates
great/reliable data. Solido is a great vendor to work with; they always
listen to the customer and provides excellent support.
I would definitely recommend it.
---- ---- ---- ---- ---- ---- ----
Solido Variation Designer
We use Mentor/Solido Variation Designer to check design variations and
get design yield estimation based on our SPICE simulations.
I estimate Solido gives us at least a 50% savings in time and effort.
Their accuracy is good, also. We use different methodologies of the tool
i.e. High Sigma Monte Carlo (HSMC) vs. High Sigma Verifier (HSV) -- and
the results align with each other.
Solido's data visualization is good, though it has room to improve for
cases where the results do not converge (e.g. due to not enough sample
points), as the data visualization is less friendly for those cases.
I have a very good impression of Solido Variation Designer.
Even after the Siemens acquisition, they still have excellent customer
support; they respond to my issues and get them solved quickly.
---- ---- ---- ---- ---- ---- ----
Solido PVTMC Verifier
We use Mentor/Solido PVTMC Verifier to check variations at multiple PVT
corners.
I have a very good impression of the tool.
---- ---- ---- ---- ---- ---- ----
Mentor-Solido Variation Designer
It does corner case variation and Monte Carlo analysis for your circuit
design, such as random local as well as global mismatch.
The problem with doing Monte Carlo analysis *without* Solido is that you
must run a lot of simulations to get a Gaussian distribution to
characterize if the circuit will be functional or not -- in light of
realistic and unavoidable device variation.
Solido lets us target the distribution tails we need to understand
those extremes. You put in the circuit schematic that you want to
simulate, along with the device elements and their random variables.
Rather than picking a random number (i.e. a Monte Carlo approach),
Solido populates the variables more cleverly. Just think of how many
dice you'd have to roll, or in this case how many simulations you'd have
to run, to construct the distribution at the tail.
Examples:
- 3-sigma. The 3-sigma "tails" from the mean are each ~0.17% of
the population. So, + or - 3 sigma is 2*0.17% or 0.34% of the
total population. That is about 1 in 300 events (i.e. 300
is the reciprocal of 0.34%).
This means that to catch a 3-sigma failure, you must run 300
simulations if you only used conventional HSPICE or Spectre
simulators -- i.e. without Solido.
That's generally not such a big deal. However, when you need
higher margins, e.g. the gaussian/bell curve is even further out,
so events with lower probability.
- 6-sigma. You might have a memory array which typically consists
of way more than 300 cells, usually in the millions.
You don't want to see a single failure. So, 6-sigma
compliance means no failures out of 1 billion, so you'd have
to run a billion simulations.
This is where Solido adds high value. Solido cleverly selects
the right simulations, so you only need to run several 1,000
simulations -- instead of 1,000,000,000.
We work closely with Solido. Their customer support is a significant
part of their strength. They get back to you quickly to resolve issues
with designer usage and tool issues. They want to make sure you are
using the tool productively so that you are hooked on using it.
Solido has put together a good product with Variation Designer.
---- ---- ---- ---- ---- ---- ----
Solido Variation Designer
Solido Variation Designer is a variation design tool that helps reduce
the number of Monte Carlo simulations and provides insights to worst
case corners.
I did a trial use of it for a 3-sigma design -- it dropped the number of
Monte Carlo simulations I needed from 300 down to 50.
Since it was for post-layout, I haven't done a direct accuracy check
yet. (You could do it more easily for a schematic.
One nice feature Solido is it shows which parameter is impacting the
design variation the most.
---- ---- ---- ---- ---- ---- ----
I use Solido Variation Designer with Mentor ELDO simulator for our
4-sigma designs.
I used it to
1) ensure output matching in a current reference block, and
2) to see if I had the right trim capability.
Solido allows you to have multiple testbenches.
- It let me use the outputs from my previous testbenches as
input to the next ones.
- For example, it used the output from testbench #1 as the input
to testbench #2, etc... So, I was able to have continuous
serial runs.
I liked it a lot -- it helped me out.
---- ---- ---- ---- ---- ---- ----
Solido Variation Designer is a good tool for running a large amount
of Monte Carlo simulations at a fast speed.
Its interface is easy to use; even without training. People should
be able to easily figure out how to use its functions.
Solido is pretty helpful. Our company has an issue of limited disk
space, so the large amount of files that Solido generates was a
challenge. Spectre took longer for the Monte Carlo simulations, but
used fewer files, so our database could handle it better.
---- ---- ---- ---- ---- ---- ----
Solido PVTMC Verifier
Solido PVTMC verifier is a nice tool that provides insight on your
worst-case corners, and with a good graph.
I've tested it out. It gives you the worst case corners from what you
pre-select, i.e. different temperature corners and different extraction
corners.
It's a useful tool, but at the end of the day, the designer must still
know the design to get accurate results from it.
---- ---- ---- ---- ---- ---- ----
Related Articles
User buzz on Siemens/Solido machine learning is #1 for Best of 2018
Solido ML, BDA Ravi, Tom Beckley SSS makes #2 for Best of 2017
BDA, Solido, MunEDA, and Silvaco get #5 for Best EDA of 2016
Mentor BDA AnalogFastSpice and Solido were #4 tools at DAC'15
Join
Index
Next->Item
|
|
