EDA Embraces Big Data Amid Talent Crunch

The semiconductor industry’s labor crunch finally has convinced chip designers to bet big money on big data.

As recently as 2016, executives weren’t sure there was a market for big data approaches to electronic design automation. The following year, utilization of big data remained stuck in its infancy. And in 2018, Semiconductor Engineering questioned why the EDA sector wasn’t investing in big data and machine learning.

Since then, chip design has grown more complex and data-intensive, and many organizations have been struggling to find enough engineers. To compensate for that talent shortfall, the EDA sector has bet millions of dollars on products and services aimed at helping fewer design engineers complete more work, often through the use of artificial intelligence that is fed vast quantities of data.

AI can find patterns and use them effectively at speeds and volumes far greater than their human counterparts. Doing so requires harnessing big data and using techniques like machine learning and deep learning to train the AI to detect the relevant patterns. This process frequently requires a sophisticated hardware design that is time-intensive for human engineers. Ironically, those same designs can be used to fuel the productivity tools that can make the design process more efficient. In other words, AI creates a problem when good engineers are hard to find, and also serves as a potential solution.

Humans are still an essential part of the chip design process. The industry’s long-standing concerns about a labor shortage have been exacerbated as of late by tech and auto giants competing for the same candidates. At the same time, the slowdown in Moore’s Law means engineers have to develop more complex solutions to improve chip performance, instead of simply shrinking down the components, a process that requires increasingly niche skill sets from engineers themselves. And all of this is happening amid a global chip and wafer shortage, while unprecedented demand drives the industry toward $1 trillion in revenue by the end of the decade.

Several companies say big data can help by making digital design engineers more productive. At the same time, industry advocates argue that resolving labor issues in the long term will require systemic changes that technology can’t address.

Still, Mark Richards, a senior product marketing manager at Synopsys, noted that productivity gains can create meaningful change over the right timeline.

“It’s like compound interest in your bank account,” said Richards. “Even small efficiencies can compound over time, especially if you can make every engineer at the company more efficient. Then you can start to address those holes around not having enough engineers or not having enough experienced engineers. A lot of the macroeconomic changes are going to take 20 years to come out of the pipeline. This kind of technology can make everyone more effective today and help the shortfall that’s happening now.”

However, Rupert Baines, chief marketing officer at Codasip said it is not about the data. “It’ about more productive tools. Data is inert, so you need people to process data to tell you something: information. But, if you simply don’t have enough people, you need smarter tools. Some tools can be statistical and analyze the data. But some can be used upfront to make the development more efficient.”

Baines noted it is striking that the software world did not have the same talent crisis. “The software world has increased their use of more efficient tools and managed it in a way hardware has not. 20 years ago people write code on C and assembler. Now they use richer languages, like Haskell, Scaka, Go or Rust. 20 years ago people also designed chips in Verilog. And they still do. System Verilog is an advancement, but a minor one.”

He believes a new generation of tools and languages is needed like the software world has, to increase efficiency and relieve the talent shortage. “Data on its own will not help. It is easy to drown in dumb data when actually we need to move in the opposite direction. Move up the stack, with tools that gives us knowledge. As T.S. Eliot is often quoted, ‘Where is the Life we have lost in living? Where is the wisdom we have lost in knowledge? Where is the knowledge we have lost in information?'”

Roddy Urquhart, senior director of Marketing at Codasip said this is why RTL synthesis was a breakthrough in the early 90s and processor design automation is today. “The other thing with data is that it is the quality of the data rather than the quantity that matters. When thinking of the verification of processors people tend to think that the volume of test vectors is the only thing that matters. However with random testing this may be counterproductive.”

Design is a party, design is a car
What does EDA have to do with an American pizza restaurant chain with an anthropomorphic mouse mascot? Richards said that, at their best, digital design tools for big data are the equivalent of a one-stop shop for children’s birthday parties. “At these places, you have everything you need to have a great party — the games, the plates, the tables. You can just have fun, and do what you need to do. Similarly, this is an out-of-the-box way of doing design.”

Fig. 1: A summary of the difficulties facing the EDA industry according to Synopsys customers. Source: Synopsys

Just as Chuck E. Cheese wouldn’t exist without exhausted parents looking for an easy way to host an event, Richards said Synopsys has had several industry challenges in mind as it develops new products. First, there was the concern about an increasing shortfall in designer productivity as designs and systems grow more complex. Second, the opacity of the design process makes it difficult to track and improve. Finally, developing solutions to these issues in-house requires even more resources and distracts from the main task the company is trying to accomplish.

The answer to those issues, according to Richards, involved bringing together big data, machine learning, AI and analytics technologies to create a “machine intelligence-guided way of doing design. We’re trying to make engineers smarter by giving them the data they need to make decisions more effectively. We take the big data that is generated by all the different point tools, and find all sorts of things that engineers themselves can’t find very effectively because they don’t have time to find those patterns. You can now say, ‘the design team tends to get stuck for six weeks and it all came down to RTL.’ It starts to give you big ideas about how to improve the process.”

Richards says this type of technology has interesting implications for areas like automotive and space exploration.

“Data-driven decision making is where people are going,” he said. “You might find out in five years that your data has massive value, but you won’t know unless you save it. Ultimately, this will affect much bigger systems with much bigger dollar amounts on the line than just making a little IoT device a little faster. There’s an opportunity to make bigger systems safer and more reliable.”

Kam Kittrell, senior product management group director in the Digital & Signoff Group at Cadence, said he thinks about AI design tools in terms of cars. At one end of the spectrum are fully manual autos. At the other are completely autonomous vehicles that deliver the passengers safely to their destination with no driver assistance along the way. “Where technologies get exciting is toward the autonomous end of the spectrum, with conditional and high automation. Once you give the parameters, they can run hundreds of machines at a time and hundreds of experiments at a time, whereas humans can only do two or three.”

With big data fueling such productivity gains, it may seem odd that the industry didn’t anticipate the role it would play in EDA several years ago. Kittrell said that may have to do with an underestimation of the possibilities of AI.

“A lot of times people were thinking about artificial intelligence in terms of picture recognition,” he noted. “As in, ‘Here are 12 pictures, tell me which one is a cat.’ And the more pictures you give it, the higher the efficacy of determining which one is a cat. The principles of artificial intelligence are about 60 years old. But it was only in the last 10 years or so that compute power and storage caught up with the concepts in order to make meaningful neural networks that have enough nodes, and so forth, to do something interesting.”

On the other hand, the role of such technologies in the short term also can be overestimated. Kittrell said these tools are unlikely to completely replace human engineers working on such complicated problems any time soon.

“The semiconductor industry is going through a stage of unprecedented growth, requiring large teams of engineers to keep up with demand,” he said. “There have been four strong growth vectors in the past five or six years, which are all converging. There’s mobile, which has been around for a while, but expanded with 5G. Automotive is growing with self-driving car technology. There are hyperscalers like Amazon using very specific, customized silicon that will be replicated throughout all their servers in all their data centers. And then there’s AI, which is more power-efficient, and faster if you build specialized hardware.”

So, will the productivity tools that develop out of necessity during this growth period put human designers out of a job? Consensus says this is not likely.

“We’re going to need more humans,” said Kittrell. “They’re just going to be doing different things than what they were doing before.”

Building the talent pool
Experts across the industry and academia agree that big data alone is unlikely to resolve longstanding labor issues. One key reason is that tools advanced enough to solve sophisticated problems for engineers must themselves be designed by teams with a sophisticated understanding of the issues at play.

Whether the sector is EDA or something else, other commonly-cited solutions to the industry’s labor issues include a commitment to diverse and inclusive hiring practices, an increase in wages, and educational and training investments. Santosh Kurinec, a professor at the Rochester Institute of Technology, said a comprehensive solution must include educational enhancements in K-12 and beyond, as well as “re-skilling” or “up-skilling” current employees.

“The semiconductor industries rely on highly educated and experienced electrical engineers, chemical engineers, physics engineers, mechanical engineers, and software engineers to continue pushing the envelope of semiconductor chips, whether it’s in process technologies or chip design,” Kurinec said.

Bob Smith, executive director of the SEMI ESD Alliance, noted that while analytical tools AI and machine learning are helping designers manage complexity, the complexity in turn is continuing to grow.

“I would liken this to a race where these tools are helping designers stay in the game, but not reducing the need for more designers,” said Smith. “In fact, the tools need expert designers to understand how and when to apply them, and to realize when to make course corrections. The big picture is that advanced tools are very necessary but do not lessen the need to continue to build the talent pool.”

Mark da Silva, senior director of the SEMI Smart Manufacturing Initiative, agrees with Smith that such tools are “helping and necessary,” but they only go so far. “EDA plus AI can potentially provide additional paths to improving chip performance,” he said. “Designers and companies that can harness these techniques will reap benefits of improved chip performance by squeezing out more for a given design and doing it faster. Education and training related aspects for the workforce will potentially be more critical.”

It’s easy to get lost in a human-vs.-machine debate. But the big picture on big data is that, when combined with AI, it can help augment human decision-making, noted Matthew Hogan, product director at Siemens Digital Industries Software.

“One of the things we’re seeing with AI and machine learning is that the parameters you need to put into these systems are very specific,” said Hogan. “Part of the challenge I personally see for this automation is how to quantify the experience, or what we’ll call the gaps in the specifications, like you didn’t specify a pin voltage for something. But it’s evident to a designer, either on the analog or digital side. All of those learning attributes that you pick up from being a new grad, to being a newbie on projects and being experienced and confident, will be very difficult to translate into an automated design environment. There’s still going to be the need for the design engineer as functional oversight, or if you like, a purveyor of goodness to be able to guide and direct those systems as we continue to go forward. It’s often the nuances that make the biggest difference.”