How Cloud-Based CAE Is Changing Prototyping: An Interview with Ian Campbell, CEO of OnScale
The days of manual prototyping may be numbered, but so could prohibitively expensive IP licenses.
Traditional prototyping methods are evolving. The advent of complex designs such as ultrasonic sensors has raised the bar for simulations. OnScale CEO Ian Campbell says they've created cloud-based CAE (computer-aided engineering) software to make this kind of difficult design more attainable and affordable—and put a shelf life on the days of in-house simulation in the process.
If you were to ask an engineer where the biggest challenge lay in designing new products, you would get back a variety of responses. However, chances are good that many responses would involve the challenge of balancing the scope, timeline, and cost of prototyping.
There is a variety of computer-aided engineering (CAE) software programs that have long helped engineers with bringing their designs to market. Existing software requires multiple licenses, without which design teams find their work bottlenecked.
Ian Campbell, CEO of OnScale, believes that there is another way: cloud-based CAE, which enables hundreds of design variances to run through simulations simultaneously, without requiring engineers to bear the financial brunt of acquiring full program licenses.
AAC recently met with Campbell to discuss how cloud-based CAE works, examples of it in action, and how OnScale aims to revolutionize the prototyping process.
All images courtesy of OnScale
AAC: Tell us a little about your background and how your own experiences helped drive the beginning of OnScale.
Ian Campbell: I'm a mechanical and aerospace engineer by training, but I kind of became an electrical engineer through my first startup, NextInput, which made IoT force sensors for 3D touch, force touch applications for smartphones, wearables, automotive, IoT devices—all sorts of different applications.
We used computer-aided engineering (CAE) tools to design and optimize our sensors. We kind of had to at that stage of the company because it was very expensive to go and tape out one of these sensor runs and actually fab the sensor with a foundry partner. In some cases, it would cost upwards of a million dollars to get a single [way] for our prototype sensors.
There’s no worse feeling than when you spend that amount of money and then, when you get your engineering samples back, they don't work. That's one of the worst experiences of my life, actually.
OnScale is trying to solve that problem by encouraging engineers to use more computer-aided engineering, use the cloud’s infinite computer resources to take advantage of the infinite computing resources to design and optimize your devices before you go spend a lot of money prototyping things. We really think [CAE] is the future of engineering and we're going to disrupt a $10 billion industry.
There's no worse feeling than when you spend that amount of money and then, when you get your engineering samples back, they don't work.
AAC: Can you give us a rundown of what computer-aided engineering is and what need it fulfills in the design process?
Campbell: Computer-aided engineering uses software tools to tell engineers how to create the best design.
So let's say we're engineers and we have an idea. We would design our idea using CAD. Then we'd simulate the design using CAE multiphysics solvers. In some cases, if it’s a very complex design, that can take months. These are very computationally intensive simulations. Once we get our data back, we analyze the results and do that over and over and over and over again.
An optimization plan for determining simulations regarding the metalization ratio of a SAW filter
Traditionally, this has been a very serial process. We’re doing all of this on a local laptop or a local high-performance computer.
The problem—or the opportunity, really—is that there are 100 billion core hours used per year by engineers to design and engineer new products. Core hours is how we measure computational power. For instance, if you're sitting there with a laptop that's a four-core laptop, and you read for an hour, you've earned four core hours. 99% of it is on-premise, meaning it's not on the cloud today. Engineers are doing this design with local laptops, local HPCs. If you take a look at any engineering firm like Apple or Samsung or NextInput, we get these bursts of engineering workloads.
Think of CAE as the addition of software like Comsol and Ansys or, in the EDA space, software like Mentor Graphics and Cadence. Those are all very expensive licenses and they've also got very expensive computers that are running them. And the computers plus the software licenses are your CAE supply—a fixed CAE supply, meaning you have a fixed number of software licenses and you get a fixed number of computers.
In periods where your demand is far exceeding your supply, your risk of wasting R&D time increases. Different team members could potentially be sitting around waiting for data or waiting for access to the CAE systems. This means you’re also increasing risk, meaning engineers aren't getting the data they need to make informed design decisions and could make suboptimal decisions that lead to poorly performing products in the marketplace.
And then, conversely, if your supply is less or your demand is less than your supply, you're just wasting your development budget. All those expensive licenses and computers are sitting around not doing anything. So, fundamentally, legacy CAE does not scale.
AAC: How does OnScale address all of these design and prototyping hurdles?
Campbell: At OnScale, we're fixing these hurdles by doing CAE in the cloud.
You can download our free software that runs on Macs, PCs, or tablets. There are no licenses; there's no IT setup; there's really no hardware. You can run our program on a very stripped down laptop. And when you're ready to do one of the mathematically or computationally intensive portions of your simulation, you send us the files of the engineering problem, and we create the hardware infrastructure for you, on AWS (Amazon Web Services) or Google Cloud, or even on your private cloud if you have a big HPC cluster on-premise in your company.
Then OnScale creates just enough virtual hardware needed to solve that problem in the most cost- and time-efficient way. And we send you the data back, sometimes 100 times faster than you can get with legacy CAE solutions running in sort of a serial fashion. This is great for massive design space exploration and really tough engineering problems. We really think that we're going to help engineers bring awesome new technologies to market faster.
AAC: You used the term “virtual hardware”. Can you give us a succinct definition of what that is?
Campbell: Virtual hardware means that we look at the engineering problem in terms of the RAM and the core hours that are required to solve that problem. From that, we can create what are called virtual machines on AWS using what I call containers, or dockers, which can be of any given size.
Say your problem required 32 hours to solve and a terabyte of RAM. What we can create is a virtual machine with 32 cores and has a terabyte of RAM and we can solve that for you in one hour. In comparison, if you had a four-core machine, it would take you eight hours. And you may not even be able to solve the problem if you run out of RAM on your local four-core machine.
OnScale is trying to abstract away the hardware layer. In fact, engineers shouldn't have to worry about what happens beyond the point of setting up their models and then running an estimate and seeing how many core hours it's going to take. That is how we charge for the service: we charge based on usage. You only pay for what you use, for the core hours that you use.
AAC: You mention that OnScale is trying to “abstract the hardware layer away”. That’s a revolutionary concept when you're talking to hardware designers. How do you, as an engineer, view this huge shift in the way engineers work?
Campbell: I think that this is as groundbreaking as moving from an abacus to a calculator or from 2D CAD to 3D CAD. It is going to fundamentally change how engineers engineer. I suspect that in several years you’re going to be thought of as a dinosaur if you have your own computer hardware for engineering software locally on premise at your company. Everybody's going to want to be doing this in the cloud.
For example, our program can run 400 different design variances at once, and each design has a small change in cell radius, thickness, layers, and so on. The 400 variances all run simultaneously, getting the data back much, much faster than if we had actually run these 400 in series. We get a ton of data back and we get to see how things perform in 3D.
Screenshot of an example process optimizing a 5G RF filter
That's something that engineers have never had the capability to do before. Traditionally, if you're going to do 400 simultaneous simulations using a legacy CAE vendor like Comsol or Ansys, you have to have 400 licenses—in some cases, that would be millions of dollars worth of software. But we just did 400 simulations parallel in the cloud without any license. If I were to pay for [this simulation on the cloud], it would be maybe a couple thousand dollars to run this massive simulation design sweep.
AAC: How do you handle a program this complex? To run that many simulations and be accurate across the board is a huge task. Even a small problem becomes a big problem very quickly across that many simulations, right?
Campbell: Yes, true. It's the old adage—garbage in, garbage out—regarding computer engineering tools. We’re working on ways to filter those out. For example, if the engineer ran these 400 simulations, they set up this parametric sweep and they have values or permutations that are not going to work, we can filter those out and tell the engineer, "Hey, you're trying to do something that is impossible to do in reality, so don't even simulate it."
It's the old adage—garbage in, garbage out, regarding computer engineering tools.
AAC: What have been the biggest challenges in the development of these tools?
Campbell: The tools are based on multiphysics solvers, multiphysics algorithms that have been developed over 30 years. So this was not only developed but also validated over 30 years. It's not like we started OnScale in 2017 and all of a sudden we have these awesome multiphysics solvers that no one else has.
We inherited these tools from one of the biggest engineering consulting firms in the world who developed these multiphysics solvers for DARPA projects, DOD projects, and military projects. And then that firm stopped doing defense consulting and we got all of the really awesome, super efficient, multiphysics solvers that weren't developed from the cloud, they were developed for highly parallel mainframe computers. It just so happens that it's a perfect match for the cloud. So we decided to take them, put them on the cloud, and deploy them so that any engineer can access them.
That’s where the genesis of this technology occurred. The challenge for us was building an intelligence hardware provisioning system for the cloud where we can look at an engineering model and determine how many core hours it is going to use, how much RAM it is going to use and create little containers on the cloud that will run each individual problem separately.
AAC: Anytime the idea of using the cloud for something traditionally protected by a secure closed network comes about, the question of security comes up. If people are sending you their intellectual property to build and prototype their design, how do you ensure security?
Campbell: Frankly, security is the biggest bottleneck, or roadblock, for us today. However, I don't think it will be a roadblock in two or three years, and here's why.
We built our program entirely using new AWS services that no OnScale employee could ever actually see any customer IP. The only way that anyone else could hack the customer’s IP is if they were able to get the encryption keys that the customer generates.
Just to walk you through what this means, if you download our clients, you get a piece of software that runs on your local hardware. The first thing you do is sign on and create a Cognito account, which is an AWS service that stores your username and password. This makes it so the username and password are not stored on any system that OnScale manages, this is within AWS. So if you lose your password or your encryption key, you don’t contact OnScale, you contact AWS.
AWS’ s key management system creates a unique encryption key that's tied to your hardware. You have to keep that or else you won't be able to decrypt the data that you get back from us. And if you lose your key, you have to go through this process again to create a new key and you'll lose any encrypted data that's still in the cloud that you haven't pulled out.
When you go to run a problem, Lambda creates the computer resources needed to create these containers. You send an encrypted file up to S3, which is Amazon's Storage bucket. Then only the container that has the encryption key that you passed into can open up and decrypt that file.
AWS processes the data using our solvers. We put ourselves into these containers and the data is stored back into S3, encrypted. And then your computer can go grab it and decrypt it locally. So it's 100% on AWS and AWS has data centers in every region of the world.
The biggest takeaway about security is this: OnScale never has access to any of your IP. The only way that any one can hack your IP is if they physically go into an AWS data center, repel down ninja-style into the data center, and try to find the computer that's running your problem out of the millions of computers that are in the data center. From there, they’d need to do a RAM dump to pull out your IP. That's the only way they can get it. It is more secure than local data centers.
The biggest takeaway about security is this: OnScale never has access to any of your IP. The only way that any one can hack your IP is if they physically go into an AWS data center, repel down ninja-style into the data center, and try to find the computer that's running your problem out of the millions of computers that are in the data center.
AAC: Can you give us an example of CAE in action?
Campbell: A great example is ultrasonic fingerprint sensors, because it's the best biometric sensing technology. Ultrasonic fingerprint sensors can go underneath a display, enabling edge-to-edge displays in products like smartphones. It’s able to perform 3D transdermal sensing, looking at the 3D peaks and valleys of your fingerprint.
The challenge is it's very, very complex. You have MEMS transducers on an array, on a silicon die, numbering into the thousands.
The way I like to describe this is it’s similar to a medical ultrasound transducer that you would use to look at a baby in a womb. However, imagine shrinking those transducers and putting thousands of them onto a little piece of silicon that measures five by five millimeters, with each transducer measuring only a few hundred microns across — there’s the engineering challenge. With legacy CAE, engineers can't simulate this. They can't do massive optimization studies to optimize the individual cells. They can't do a simulation of that entire die with thousands of transducers working together.
But they need to test the algorithms, which they can’t do without making prototypes. Traditionally, in order to build the algorithms, they make thousands of different phone samples, or IoT device samples to see how each die performs when it's actually in a system. And then they create AI algorithms from the resulting data. This prototyping process costs literally millions of dollars just to get the data to make the AI algorithms that power these things.
However, OnScale can do a full 3D simulation of the die that creates ultrasonic waves that go up through material stack up of a smartphone or IoT device.
Then we can rapidly test and perfect those AI algorithms. Instead of building thousands of physical prototypes, we can build thousands of virtual prototypes computationally and analyze the performance of various parameters of the design. We can look at different material stack ups and see how those stack ups change mechanically over time, in different temperatures, or from manufacturing variances. From there, we can help the engineers make very robust algorithms that can identify a fingerprint perfectly across different devices.
An example fingerprint scan from an ultrasonic sensor
That's a huge time saver for engineers. They don't have to spend millions of dollars making prototypes; they instead do it virtually using the power of cloud CAE.
It's amazing technology, but when you're really pushing the envelope in technology, you need a ton of engineering tools to make sure you're getting it right. The concept of fingerprint sensing goes back to the early 2010s. So around 2011, 2012, people started publishing papers around the fundamental technology. And it's taken close to 10 years to actually get it into a smartphone and they aren't even in smartphones yet. Apple wanted to put one into the iPhone X, but couldn't get it to work. Samsung wanted to put one of these into the Galaxy 9, but ended up pulling it out and putting the fingerprint sensor on the back of the phone, because they couldn't figure out how to put an ultrasonic fingerprint sensor edge to edge under the display.
It’s a serious pain point — something you can only solve with a computer-aided engineering. You can't solve this problem by building tons of physical prototypes, wasting a billion dollars in without ever getting to the solution. By using CAE, you can really reduce the risk in terms of budgeting and getting to market faster.
AAC: Aside from being used in smartphones, where else do you anticipate seeing ultrasonic fingerprint sensors?
Campbell: These ultrasonic fingerprint sensors are going to be in laptops, potentially even Alexa-type devices. They're going to be in smart door locks. It’s going to be on automotive applications because ultrasonic fingerprint sensors work through water and are weatherproof.
I believe the analysts say that it'll be a $20 billion market very soon.
The reason this is so interesting to the people that are designing fingerprint sensors is because they’re arguably the most complex sensors ever developed. And I say that as a sensor guy.
AAC: What’s the main takeaway you want to leave us with regarding CAE and cloud-based prototyping?
Campbell: OnScale wants to encourage engineers to take a closer look at the cloud and cloud computing to solve really tough engineering problems to avoid wasting as much money in making physical prototypes. Instead utilize OnScale’s CAE so that when you go to make physical prototypes based on your results, you have a higher likelihood of those prototypes working, leading to design wins.
Thank you for your time, Ian!
If you have experience with ultrasonic sensor development or cloud-based CAE programs—or if you have something to say about how engineering's changing over time—please share your thoughts in the comments below.