One-On-One: Walid Abu-Hadba

Walid Abu-Hadba, chief product officer at Ansys (and a former top executive at Microsoft), sat down with Semiconductor Engineering to talk about systems engineering and why the starting point is no longer the SoC. What follows are excerpts of that conversation.

SE: How do you define system?

Abu-Hadba: It’s everything. It’s the entire product and where the product lives. The way we look at it is not from the chip up. It’s from the system down. It’s a different way of thinking about it. So if you take a jet engine, what’s the system? It’s the whole jet engine. It’s not the component or fluids or the structural part. It’s all of that plus the electronic systems and the sensors. If you look a drone, the whole drone is the system—the rotating parts, the electromagnetic parts, the chips. Then you dive deep into the physics of each. The way EDA has been looking at it is from the SoC. But an SoC sits in a phone or any form factor. So how does that SoC interact with everything around it? If you don’t think of it as a system, you’re making a mistake.

SE: Is the SoC the most complicated part of the system?

Abu-Hadba: No, the system is the most complicated part of the system. It’s the interaction of the parts. That’s not to trivialize the SoC, because it’s very complicated. But what is the behavior of the end product? We had a customer come in that builds sensors. Their sensors are supposed to last for two years. These are mission-critical sensors. But the sensors were failing rapidly—about 80% at two months. The real issue is that you can never test every real-world possibility. If you take this sensor and put it in the North Sea, where it’s cold and windy, and then you put that same sensor in an oil field in the desert, it’s a totally different environment. But it has to work. It’s not about the cost of developing the sensor. It’s the reputation, the product promise, the quality, and all of these elements.

SE: So what was the problem?

Abu-Hadba: There were a multitude of problems. This is where systems come into play. There is no one single failure point. That’s the best thing about simulation. They did a good job on the laminates expecting a salinity rate of X amount. The salinity in the Arabian Sea and the North Sea are different. That’s one element. There’s also a difference in the temperature of the water. When you put something in water that’s 52 degrees and water that’s 82 degrees, it changes the behavior of that device structurally. The material they used starts degrading much quicker. It’s structural and fluid engineering together. If you used only one set of physics, you wouldn’t have been able to predict the outcome. You had to use multi-physics. It has to be simulated in those kinds of parameters.

SE: Could they afford it?

Abu-Hadba: Yes, the manufacturer is a big company, so they can do a lot of testing. But we have another customer that’s a startup. They are developing gears. I asked what would happen if they used that same product in Dubai and Iceland. They looked at me and said, ‘We have no idea.’ You need to simulate this. It’s like asking whether it makes a difference if you have a football that’s not fully inflated? We looked at that and said the difference is too small to be noticeable. But it does make a big difference with a soccer ball on the feel of how you curve the ball.

SE: One of the problems with simulation has always been the resources need to run it. How do you simulate an entire system?

Abu-Hadba: That’s correct, and you can’t solve the problem in one dimension. Part of the solution is the improvement of computing technology. The cloud will give us a significant amount of enhancement. So will the GPU. It’s almost linear scalability. It’s almost impossible today for a small company to buy a supercomputer or 100,000 cores. But Amazon, Google and Microsoft will allow implementations of this kind of solution in the cloud.

SE: Cloud technology hasn’t fared well in EDA historically, because everyone was concerned about their IP being compromised. Is that shifting?

Abu-Hadba: Yes, and we’re seeing a tremendous change there. People are saying they need burst mode. First of all, the No. 1 platform for high-performance computing is the cloud. High-performance computing and the cloud are synonymous. It’s the biggest usage of cores.

SE: The big companies have done their own internal clouds.

Abu-Hadba: Yes, but even the biggest companies on the planet still don’t have enough compute power because of the way IT constraints work. They have massive supercomputers, and their No. 1 request from a small department is to use Amazon for this type of work. They send it out and get the results back. For industrial design, security and confidentiality is incredibly important. With the new cloud models coming up from Amazon, Microsoft and Google, this problem is being resolved now. We have one customer that is in beta with Amazon right now with their own private cloud. They’re using a lot of cores—huge numbers—and it’s cheaper and easier for them to do it that way rather than on-site. Calling the IT department and having them create something from scratch is too expensive.

SE: There’s a whole new computer architecture being used, right? It used to be commodity blade servers. Now it’s customized configurations.

Abu-Hadba: Yes. People are talking about ARM architecture servers in there. And the GPU has advanced significantly. We just did some work in our mechanical and fluid servers that we improved by 5X. The ability to use GPU and CPU together has changed things. In the structural space we were told to do 1 billion degrees of freedom. It was impossible. Three or four years ago we had Ph.D.’s who told us it couldn’t be done, and all you need to do to challenge software engineers is tell them something can’t be done. Last year we were able to solve that on a 64-core machine.

SE: What does 1 billion degrees of freedom mean?

Abu-Hadba: You can simulate an entire turbo machine in one run instead of dissecting the problem.

SE: How much data is that?

Abu-Hadba: It’s hundreds of gigabytes, and we have to chop up that data and send it back and forth to memory. It’s a lot of data. It also depends on how coarse you want your mesh and how fine you want your answer. If you are an industrial customer, good enough is never good enough. You want it as close as possible to reality. Sometimes, if you’re doing just a quick analysis, you want to see if the stresses are working, you change the coarseness of the mesh and get a result within minutes. It’s directionally right. But when you’re flying a plane, directionally right doesn’t work. We have simulation runs that have been running continuously for six months. The EDA space is very small compared to the fluid space. The fluid space is all differential equations and matrices and you’re trying to solve them. In Europe they wanted to simulate the body of the frame. They ran the simulation for 52 days, but they got incredible results. That sounds like a lot, but it’s a massive improvement. Our goal is to get it to five days or five hours. We’re never going to stop.

SE: So how does this all come together?

Abu-Hadba: You take it from the electrons all the way to the fluids. The tools are very difficult to bring together and they’re very much in isolation. We need a workflow where the fluid engineer, the electrical engineer and the structural engineer see the same results. Sometimes a structural engineer comes in and says he wants to increase a hole by 2 millimeters because it makes the compression more efficient. But they haven’t thought about what that will do to electrical interference.

SE: Going back to the cloud, the cloud is now part of the entire system being created. We use the term IoT, but that’s a superficial label for how that affects design.

Abu-Hadba: Terms will change. Five years ago we didn’t call it . Brilliant marketing people will find some other term that’s sexy in the future. For us, it boils down to the following construct: Everything on the planet will be connected with a sensor. I’m not making a judgment call about what the sensor will look like or what it will use as a communication protocol. But I know for a fact that we will be communicating through sensors. The lights in my house in 10 years will not be the same as they are today. It won’t be the same for every house on the planet, of course, but it will be some of them and that will be the trend. And I know 10 years from now that when my youngest son is driving that the car he’s using will avoid accidents as much as possible.

SE: So what does that mean from a tools perspective?

Abu-Hadba: You’re going to need to view this in four dimensions. First, you’re going to need an antenna. Everything will communicate through an antenna. The antennas will not be flat antennas. They’re going to be all different shapes. We solved the bolt problem on antennas, which isn’t trivial. A lot of people couldn’t simulate the bolt. It’s a big problem because it involves the strength of materials. An antenna could be under the ocean, in space, or in the desert. It could be wearable. Antennas will allow people to collaborate. There will be a lot of spin-offs of just antenna companies. They will be designing antennas for parts. We’re going to have high-power, high-gain and low-gain antennas. There will be a lot of antennas in cars and other protocols. In 10 years, we’ll probably have three or four magnitudes more antennas.

SE: So now that you have all these antennas, do you have to start thinking about how signals affect other antennas?

Abu-Hadba: Of course. But these antennas are not all operating on one frequency. There are antennas for GPS, for Bluetooth, for WiFi, and then there are encrypted antennas used for different purposes. A military antenna may just want to talk to one device. There will be microwaves and infrared. How do these things interfere with each other? Some of these antennas are two-way. If you’re a smart hacker, it’s one of the places to break in. If you think about it, right now we send information to the Mars Rover using an antenna.

SE: What are the others dimensions?

Abu-Hadba: The second is power. If you think about all these antennas, they’re going to need power, and changing the battery on a satellite is not an option. So everything is going to need to be power optimized. You can put a sensor and an antenna in an electric car and the power consumption of those could be higher than the motor. That’s a scary proposition, because you can have thousands of antennas in the car. You have the radio running, people charging their phone in the car. That draws off the battery off the car. Third is embedded code. A chip that does not have embedded code does not work. There are 8 million lines of code embedded on the Airbus 380 and it’s all fully certified. That’s an unbelievable number. A lot of software engineering is hard because of the different instruction sets. So you need three software engineers just to move code into a sensor. A lot of times software engineers don’t want to do that kind of work. In fact, it’s hard for industrial companies to find software engineers. They need these people for certified code because of safety issues. Last but not least is structural. When you build a sensor, you need to build it for multiple environments. All of these things are critical for the IoT, or whatever we call it.

SE: Do you have all the pieces to make this work?

Abu-Hadba: You can never have all the pieces. We have the building blocks, but there will always need to be more physics to be done. In my group alone I have about 450 Ph.D.’s. We’ll never have enough. We’ll never be done designing antennas. For power, we haven’t even scratched the surface. For structural, people either over-design or under-design. Materials cost a lot of money. You want to optimize the part. But there’s also a big shift coming. In 10 to 20 years, you won’t have parts sitting on a warehouse shelf. If you’re a car dealer, instead of getting a part from a manufacturer you’ll be able to 3D-print that part in the shop. The only way you can do it is through optimized software. That’s the difference. We’re spending a lot of time on materials science and 3D printing. If you start with CAD, you’ve missed the point. If you start with modeling and simulation, then CAD becomes documentation at the end. People today start with CAD.

SE: So all of this starts with simulation?

Abu-Hadba: John Swanson started the company and his vision was to bring finite element math to design. At that time, people didn’t understand it. It was just the beginning of the computer revolution. Fast forward a few years and the current CEO, Jim Cashman, said every engineer should use simulation in their normal day-to-day work. He didn’t say every mechanical engineer. It’s every engineer. And then he started transforming the company. He said, ‘We need to make our software easier to use. This isn’t only going to be Ph.D.’s using it. We need to make usable by every engineer.’ He then started on how to round out the physics and make the software easier to use. We then moved from mechanical engineer to fluids to electromagnetics. At the time people questioned by a mechanical engineer should be interacting with an electrical engineer. That was just as recent as 2008-2009.

SE: Is that still the case?

Abu-Hadba: The walls have broken down as companies struggle to innovate. From electromagnetics we went to chips, power and systems. We want every engineer to use tools and simulation in designing their products.