BOSTON — Engineers may worry that artificial-intelligence (AI) and machine-learning (ML) algorithms will eventually replace them as circuit designers. A panel session on June 4 at the 2019 IEEE International Microwave Symposium held here tackled that touchy subject. The opinions were generally positive for engineers, but panelists acknowledged that AI/ML will change analog/RF engineering jobs but not necessarily replace engineers.

The panelists:

  • Paul Franzon, North Carolina State University
  • Taylor Hogan, Cadence Design Systems
  • Thomas Rondeau, program manager, Defense Advanced Research Projects Agency (DARPA)
  • Sankalp Modi, MathWorks
  • Ron Rohrer, Southern Methodist University
  • Moderators: Osama Shanaa, MediaTek, and Prof. Francois Rivet, University of Bordeaux


No panelist would go so far to say that your job is 100% safe, but none felt that engineers will be able to totally coast on their expertise and intuition. Why? Because technology has changed many jobs over the years. Yes, technology destroys jobs, but it creates others. “Only 2% of the U.S. population are farmers, and we have a population that loves to eat,” said Hogan. You could say the same about manufacturing jobs. “If you’ve been doing the same job for the last five years, you’re in trouble. AI will start with the bottom jobs.” I interpret that to mean manufacturing rather than engineering or technician jobs.

“We should not view AI as an enemy,” said Modi. “The job of the mathematician didn’t go away just because of computers, though it did change. The same will happen with engineers. Unlike jobs that tend to be repetitious, engineers can use AI to produce better designs faster, but don’t expect AI to know everything you know.”

“Machines are good at dealing with more dimensions and design parameters than humans,” said Franzon. “We do a course (at NC State) mapping out the design specs, then use stochastic optimization. We can do 20 parameters and come up with a good solution.”

Modi gave some examples. “A colleague worked on a sampling filter for 1.5 months. Using CAD, it took 1.5 weeks.”

Machines can process more design parameters than humans. (Image: Martin Rowe)

Machines, however, lack the real-world knowledge of physics and math that engineers have. Indeed, panelists generally agreed that engineers can’t rely on AI/ML to simply look at how a system performs and calculate design coefficients. That became apparent in presentations at the 2019 Brooklyn 5G Summit. Profs. Tim O’Shea and Andrea Goldsmith both admitted that, while AI/ML can learn the characteristics of a wireless channel, it doesn’t inherently understand the physics behind it. But programming that knowledge into a model will make an algorithm reach results faster, with better accuracy, and use less energy.

“The physics needs to be encapsulated and tightly integrated into AI,” said Rohrer. “That is, electromagnetic effects, circuit laws, design rules, system simulation, and so on. It all can and will be integrated into AI/ML so that a designer can push on one end and immediately see the result and the specs. Or you push on a spec and see how that changes the layout.”

“We have math, we have physics,” echoed Rondeau. “AI will evolve our jobs more so than take over. ML has shown its value in image, processing, audio processing, and analytics. When we add our knowledge as engineers, we improve the effectiveness of these systems.”

“AI will not replace engineers,” said Rohrer. In the 1960s, he thought that machines would replace engineers and admitted he was wrong. Rohrer said that he was wrong then and that such thinking is still wrong.

Taking a more controversial approach, Rondeau wants computers to replace engineers, letting them go on to bigger things such as solving problems for society. “When asked if AI could replace engineers, my first reaction was, ‘I hope so. I have many problems to solve.’” Rondeau considers himself lazy by “working incredibly hard to not do the same thing twice. It’s to lighten the load on engineers and on society.”

Engineers have a creative ability missing from AI systems. (Image: Martin Rowe)