Reading time ( words)
It has been a crazy year with lots of ups and downs. But within the clouds, we’ve seen plenty of silver linings, too. Case in point: Avishtech, a brand-new company lead by founder and CEO Keshav Amla. (You may be familiar with his father, Tarun Amla, a veteran PCB materials technologist who is now with ITEQ.) We recently caught up with Keshav and asked him to discuss his company’s simulation tools, his plans for the company, and what it’s like launching a company in this “new normal.”
Andy Shaughnessy: Keshav, you recently founded Avishtech. Give us some background on yourself and tell us how you got into this industry.
Keshav Amla: I’m a mechanical engineer and a materials scientist. I studied mechanical engineering at Caltech for my undergrad. Then, I went to Stanford, where I studied materials science and engineering and earned my master’s there. While at Stanford, I started Avishtech and began working on it full-time after I left grad school. One of the first things that we wanted to focus on was the PCB industry. It’s something with which I’m familiar and an industry I felt we were well-equipped to serve. I worked a couple of summers at Isola while I was in high school. My father, Tarun Amla, has been in the industry for a long time. What drew me to engineering as a child was the fact that both my parents are engineers, so a lot of conversations would veer into the technical realm.
I was involved in those conversations from an early age. Not only did that draw me toward engineering, but given that my dad has been in this particular industry for a while, oftentimes, discussions would be specific to PCBs and laminate materials. I felt if I could make an impact in this industry and solve some key problems, that would be a good approach for me to start my company.
Shaughnessy: You’ve done research at JPL and HP.
Amla: Yes, during my undergrad, I did some research with JPL as well as at Caltech. During graduate school, I did research at Stanford, as well as some work at HP for a summer.
I’ve been fairly diverse in my approach since many different topics interest me. I’ve worked on things ranging from astrophysics and coronal mass ejections at JPL to work related to phonon transport at both Caltech and Stanford to other things that are relevant to the industry I’m in now with dielectric materials at Isola. I’ve always been very open to different aspects of science, and I’ve been a strong believer in the power of an interdisciplinary approach.
Shaughnessy: That’s pretty impressive. Your company focuses on stackup tools and field solvers. It has been a while since we’ve seen any new companies in that space. What are some of your goals?
Amla: For our first area of focus, in terms of products we’re bringing to the market, we chose two products, Gauss 2D and Gauss Stack, from our Gauss line of products. We’ve developed a tool that picks up where a lot of the other field solvers in the market leave off. A lot of things make our 2D solver unique. First, our accuracy is extremely high, largely because we use rigorous finite difference methods. We maintain a high speed of simulation that’s similar to our competition while providing higher fidelity. On top of that, we have more granularity, and we can account for dielectric anisotropy, which is something that nothing else in the market currently does.
With Gauss Stack, not only can you make complete stackups very quickly, but you can also perform an entire gamut of simulations on it. You can run an electromagnetic simulation using the inbuilt version of Gauss 2D. You can click a button and simulate all of your traces in one go, and then you also have the ability—something that nobody else has—to use our coupled thermomechanical simulation to solve for dimensional stability and warpage. As for the EM simulation, you also have the ability to perform synthesis—to obtain a trace width for a target impedance—for the entire stackup in minutes, which is another unique feature. Add that to the fact that we also allow you to get reliability estimates and have this all documented in a neat, easy-to-use interface, and we feel that we’ve made something extremely powerful.
The company’s goal is to build the materials-based technologies for the future. We feel that a lot of companies haven’t been innovating in materials as much lately. Startups tend to focus more on other fields, but the materials sector is the basis for all fundamental innovation. We think we’ve made a good foray into that and we look to do more of that in the future.
Shaughnessy: Who is the target audience for this? Are you partnering with any materials companies?
Amla: Our target market would be board shops and OEMs, or anybody involved in the PCB industry or the electronics industry; even semiconductor and materials companies will benefit from our tools. We’ve not had to partner with anybody because we wanted this to be something that you can rely on without necessarily having to be within anybody’s particular ecosystem. We wanted this to be an independent third party, not based on anybody’s particular products.
Shaughnessy: You looked at the existing tools out there and saw a need.
Amla: Exactly. We felt that without sufficient simulation tools being available, the process is very challenging for PCB designers because they’re working with limited data. We built a tool that demystifies this process to give them data and help them figure out what to do with it. Essentially, Gauss is your guide through the PCB development process. We thought not only will that result in higher performance, but also a lot less frustration on the part of engineers.
For example, consider our dimensional stability aspect. PCB shops have a tremendous amount of lost revenue and waste because they have to run scout batches. They don’t have any tool that can accurately and reliably predict dimensional deformation across layers in a board.
You end up with misalignment between the different layers of the board, and that causes issues related to registration. You’re not able to get successful registration on your first go. You have to run a scout batch and then compensate for it. Now, your board shop has not only material waste, but you’ve also lost out because now you don’t have that quick-turn capability as before. You lose money because you’ve taken more time.
Gauss Stack is a tool that can predict this for you and allow you to compensate for it in your actual development process. Now, we can drastically reduce the amount of lost revenue in waste due to scout batches and scrap. That’s one of the key aspects in which we feel Gauss Stack makes the entire industry a little bit more on track, with less trial and error necessary.
Nolan Johnson: Is there a modeling activity that the end-user is responsible for putting together?
Amla: I’m glad you asked. No, there is no aspect required from the user to model anything as pertains to both our thermomechanical and electromagnetic simulation. That’s one of the big challenges. Within the industry, most of the other 2D solvers are those of either the boundary element method or the finite element method at the higher end. Now, the finite element method-based solvers require a lot of work from the user to set up the problem. As far as building the geometry—meshing it, and then making sure that you’ve attained convergence—unless you’re an expert at FEA, you’re going to really struggle to use that software. But with Gauss 2D, you can run a problem in about 20 seconds.
You put in the inputs and have a little bit of control in that we allow you to adjust how fine of a mesh you want and how many iterations you want. But you can leave those at default to the lowest value, and we’ll still be highly accurate. If you want to increase them past that point, you move the slider up, and you’re good to go.
Johnson: Most of the engineers in this space will hear you say, “There isn’t much setup process.” Normally, your customers are going to hear that and say, “Then this must be simplistic and not very insightful as far as the results I’m going to get,” because that’s what they’ve seen with similar tools in the past. How can it be so simple to set up and still be accurate and precise?
Amla: It can be so simple to set up, while still being highly accurate, because of the way finite difference methods work. If we get down to the fundamental rule of calculus, differentiation is a finite difference method of the first order as your delta approaches zero, so there is no inherent complexity to a finite difference method. The problem is that on their own, without sufficient acceleration techniques, you end up with a very slow computation. That’s why, historically, high order finite difference methods haven’t been used for these kinds of solvers. This is what I’ve just been working on for most of my academic career. I threw in every single computational trick in the book, and we developed some more in-house.
As a result, we were able to get this to the point where not only do we have a high-order, finite difference method, but it also runs quite fast. You’re able to run simulations within just a few seconds. Now, you have something that can be used. It doesn’t require anything from the user. Perhaps most importantly, though, we have incorporated an entire spectrum of predefined transmission line geometries and combinations. In addition to standard differential/single-ended microstrip and stripline configurations with trapezoidal or rectangular traces, you can run coplanar waveguide, dual dielectric, multiconductor, and many other combinations. We’ve put in a lot of effort into making sure that this is easy to use.
As somebody who has been using simulation tools for quite some time, I know how frustrating it is that you have to keep iterating back and forth before you can finally get something that works, is stable, and gives you the kind of accuracy that you need, and that’s why we took this very different approach in our tools. This point will really hit home when our customers see the wide array of outputs and visualizations that our tools provide in contrast to the black-box numerical outputs they are more used to from other tools.
Johnson: Simple and accurate is a great benefit to me as a potential user. I expect that I am going to pay some performance prices. In a side-by-side comparison on a similar analysis, how much time does this take compared to the other ways of solving the problem?
Amla: You will not really see an impact on performance. As an example, a typical stripline configuration in Gauss 2D will take ~3 seconds on a typical desktop that our customers would run it on. More importantly, there is no setup time involved either.
Johnson: That opens it up for this application to be a more accurate solver without having to deal with the approximations. This is very interesting.
Amla: Thank you. We’ve validated this thoroughly with data that’s out there, and we are on the money every single time. With every single paper we’ve compared, we are right there. In almost every case, we’re closer than all the other solvers.
Johnson: What kind of hardware do you recommend that users have for this?
Amla: Being a computational tool, you don’t want to run this on low-end hardware. Generally, we would say that if you have a high-quality laptop, you can run this. But if you’re running this on a proper desktop configuration with good quad-core processors or better, you’ll be doing quite well. We just wouldn’t recommend using an old machine you have lying around from a decade ago. If you have a very high-end machine, you’ll see the computations happen almost immediately.
Johnson: How are you selling your software?
Amla: The software is sold as an annual subscription hosted on the user’s local machine.
Johnson: Does that mean that you can set this up as a multi-user network-based installation?
Amla: Right now, our intent is to offer node-locked, single-user configurations, but we’re willing to work with people if somebody is coming in with a particular need. We’re flexible and can work with them, and our licensing fully supports that.
Johnson: Not to put anything on your roadmap, but given the current situation with COVID-19 and the long-term prospect that we’re going to be working a lot remotely, node lock may not be the best way to go for a design team if you’re trying to break into some of the Fortune 500 design groups.
Amla: Sure. As far as the setup we have for licensing, we’re very nimble on that. We can change things up for people on an as-needed basis, too.
Shaughnessy: How big is your company?
Amla: We’ve been in a stealth mode for some time. We’re just now emerging from that and have enough people and resources to support the market. We’re looking to grow, but the COVID-19 situation is playing a bit of a role in that too.
Shaughnessy: As you said, you are launching this company in one of the craziest years in recent history. How has it affected you?
Amla: We started working on Gauss in June of last year. Six months later, suddenly, we saw the outbreak in Wuhan, and since then, things have really changed. I haven’t been going into the office since early March. As Nolan mentioned, it has been a lot of working remotely, which has been a little bit tougher. Collaboration becomes more difficult, but we’ve been adjusting the same as everybody else. As far as the decision to still go ahead, if our software and products are out there and people can purchase them, that’s, of course, preferable to being in some holding pattern.
Shaughnessy: This is a big thing for this industry. We haven’t seen a new company in this space for a long time.
Johnson: What it takes is not being a “me-too.” Be disruptive and have a very focused, clearly defined specialty. You’re doing that. You have your niche.
Shaughnessy: It has been great speaking with you, Keshav. This is a really positive thing to hear about. We so rarely see new companies in EDA. Best of luck.
Johnson: Thanks for your time.
Amla: Thank you both.