June 2, 2014, edited transcript
I work in the radio CAD group. We have a number of CAD groups here in our Broadcom, and our local group handles all the radios.
The top two challenges we see in variation-aware design:
– Design sensitivity. It’s difficult to figure out what in your design is affecting your circuit the most. We found that Solido has a very good way of being able to give us that information.
– Calibration. Another thing we use Solido for is to understand the calibration. You have steps you in calibration; you have a code that you use. And often you have to run a bunch of chips through your testing cycle in order to figure out how to linearize that. We found that with Solido we’re able to predict this much easier.
Best practices that we have – one of the things that I’ve found throughout the industry is that you tend to use models that are more digital than analog. So you get fast-fast, slow-slow corners, and so on.
Everything’s tracking up and down. Your capacitors and transistors are tracking up and down. And that’s not really real world. So when you pull out all these different effects, so your transistors are different than your bipolars, which are different than your resistors and your capacitors, you have a better chance of understanding your circuit.
Selecting cluster post processing. Solido has this ability to distribute the processing of your information. So when you run a simulation and you get your results; a lot of other tools out there only use a single node that sits there and chugs through all of your post-processing. With Solido you’re able to distribute that over your cluster.
When we first put a circuit up, we select independent sweep. That way we only run a handful of simulations. We run slow, typical and fast for each of our parameters. And that gives us an eyeball feeling of what our circuit is going to be doing.
We run a full analysis after that. But this is to let us know that we’re in the ballpark.
Our top two or three “must haves” are:
Being able to figure out what it is in your circuit that is causing your problems. It’s not always straight forward. Sometimes your p-fet is causing you a little more trouble than you thought.
In results for analysis in display: this is something that a lot of other tools don’t really combine the data in a way that’s meaningful. And we found that Solido has been able to do that for us very well.
Okay, this was kind of fun – we had a designer, that ran 34,000 simulations.
– When he didn’t select cluster-processing, it ran in over 27 hours. Because only one node was sitting there and calculating everything.
– When we turned this on we were astonished that it only took forty-four minutes. Now, a caveat to that is that it uses a huge number of licenses. So if you don’t have a lot of simulator licenses that would be a problem.
Pulling out the information, in your FETs, resistors, your capacitors, and diodes, gave us about a 20% improvement of performance over what we were seeing with our basically purely digital models.
And, that’s it.