中文 English

Steering The Semiconductor Industry

What does it take to get a new language, tool, or methodology established in the semiconductor industry? Disruption has rarely worked.

popularity

Progress in semiconductors has been one of the most successful engineering feats, and the industry has ridden an exponential curve longer than anything else in history. It is also a highly conservative industry that has pushed away many disruptive changes in favor of small incremental changes that minimize risk.

There have been significant changes over the decades, and they often required a strong voice in the industry to evangelize a new language, technique, or methodology. Unfortunately, very few people or organizations have been successful with this. The industry has done a much better job providing the necessary training and education once it has been successfully established and standardized.

There is a continuum of subjects for which training needs to be provided. They range from standard languages to disruptive methodologies. Each step requires a different approach.

A lack of evangelists
It also helps to have an evangelist. Consider Electronic System Design Automation (ESDA) or the Electronic System Level (ESL). This technology approach once had an evangelist in the late Gary Smith, but nobody has taken up the mantle since then.

Training companies help engineers master well-entrenched language and a little bit of methodology, while vendors teach engineers about unique functionality and the best methodologies to maximize the benefits of their tools. But there is a sizable gap associated with educating the industry about new methodologies or practices, especially when they involve more than one language or tool.

“Technically, Gary wasn’t an engineer or a technical expert who could develop anything,” says Simon Davidmann, founder and CEO of Imperas Software. “But he could understand the trends and understand the technology. He could see the benefits of it, and would get enthused by it. He loved to talk about it and could, in turn, enthuse other people. In that way he was a melting pot catalyst, a focus for communication. He had this respect and independence and people from the industry respected his credentials.”

Several standards bodies and consortium have attempted to fill that void, but they do not have the same access to the decision makers in organizations. They also do not share the same level of enthusiasm or the independence needed to see how many things fit together, where the holes are, the pitfalls, the weaknesses. The International Technology Roadmap for Semiconductors (ITRS) also used to perform some of that function, but since 2017 it is no longer updated. The baton was passed to IEEE’s Rebooting Computing initiative and named the International Roadmap for Devices and Systems (IRDS), but it does not appear to be having the same impact.

It may work better when a company takes up the banner for a standard. “Everything we do is based on IP-XACT IEEE 1685 standard,” says Isabelle Geday, general manager of Arteris IP Deployment. “It is our duty and our prerogative to train people, as well as we can, on the standard — its existence, its benefits, and the way to use it. By doing this, and by making the effort to do it well, we promote the standard, and long-term we promote a best methodology on the market for the next generation of SoCs. Thankfully, there is good alignment between IP providers, SoC designers, and EDA tool companies. But someone has got to promote it. Someone has to tackle the training aspect and perform the evangelization of the market.”

A lot of the time, it is new startups that are the first to push a new language, methodology or tool, and just as often they do not have large amounts of available resources. This makes it difficult to create the necessary momentum. “I see training as an ongoing learning process which allows engineers to keep on top of the changing landscape of technology,” says Ashish Darbari, CEO of Axiomise. “It is also necessary to make sure engineers always know how to learn. As engineers get comfortable, even complacent after 5 or 10 years doing the same things, they forget how to learn a new technology. Learning how to learn, asking good questions, is something very important for anyone trying to grow their knowledge. A good question is the most important first step. Those organizations that have labored management structures, with managers not open to adopting new technologies for the fear of adopting change, are the ones that lose out most in seeing benefits in a field that’s changing every six months.”

Some tool developers, who understand the reluctance of teams to adopt new things, often will make development decisions based on that. Consider the choice of language for describing RISC-V processors within Codasip. “If you look at the functional languages, it can be hard to learn because it’s a different approach,” says Zdenek Prikryl, CTO at Codasip. “C is a standard imperative language and we didn’t want to use something that was not mainstream. When we compared the pros and cons at the beginning, we found that C is well known and has enough abstractions to describe a CPU. Sometimes, revolution is good but not necessary from the language point of view. We did need to add some constructs on top of C to capture instruction encoding and the textual form of instruction.”

Training a new generation
Change can take time, and sometimes the best approach is to ensure that the new generation of engineers has the training they need to be the change makers. “As vendors, we can work with the universities to make sure that online education is available to the students,” says Bonnie Willoughby, technical education thought leader at Siemens EDA. “As they are learning from the professors and building their engineering skills, this can be enhanced by putting it to practice with real-world tools. We are investing in the future of our young generation. They are more familiar with online education, and they are more receptive to it — using that as a way to keep learning, for it not to be an event but a lifelong learning, and continue with that beyond the days of college education.”

That does not just apply to the large corporations. “We set up a university program, which now has 400 or 500 universities,” says Imperas’ Davidmann. “We are a small company and we don’t have resources to do a lot, but we try to make what we can available to help these people. They now have access to our full product suites and technologies, plus all our internal training videos.”

Not everything can be learned from a book. “Some things can be learnt using training programs and material available online, whereas others can only be gained by hands-on experience and can be enriched by learning from more senior engineers,” says Olivera Stojanovic, business development and project manager for VTool. “As an example, debug is the most challenging and least predictable process, currently without defined structure and methodology. We decided to help the verification community by establishing a debug methodology and empowering engineers by helping them master this process in a systematic way.”

Training itself also needs to innovate. “Learning could be expanded to a more collaborative platform, where every user participates in content creation, editing and consumption,” says Shivananda R Koteshwar, group director at Synopsys. “This creates an ecosystem of users who benefit from the collective knowledge on the platform. It also increases access to subject matter experts (SMEs). The semiconductor industry is built on sharing of knowledge where access to SMEs helps build a strong and robust community that goes beyond the use of any particular product.”

Training also has to be inclusive. “For online courses, we have taken great pains to make sure they are suited for people with hearing challenges, so accessibility should not be an issue,” says Axiomise’s Darbari. “For online content, I personally hand-crafted all the subtitles. For in-person courses, we take pride in taking time to understand the engineer’s requirements and delivering them what they need for their projects.”

Independent training is often highly valued. “Last year we did a seminar on methodology for verification of processors,” says Davidmann. “We explained what we see in the industry, and I showed the different open-source tools and proprietary tools from some of our partners. I’m a great believer that we need to try to help move the industry and the academics and the students forward to learn about it.”

How much should you pay for training?
There is a wide range of expectations when it comes to paying for training. This industry is somewhat different than other fields. “I was involved in ISO 26262, which is a functional safety standard for semiconductors and other electronics,” says Kurt Shuler, vice president of marketing at Arteris IP. “In that case there was an existing infrastructure for training, as well as certification companies. But when it comes to the semiconductor industry, there has to be a certain critical mass before it makes sense to invest in a Udemy course, or something like that. We think of Verilog as being commonplace, but in the big picture it is a small niche. Very few companies have figured out a business model where they can make money from that. Some make money off training, off certification, off product or company certification. They have to be creative in how they bring money in.”

Staying current on training is a significant challenge for everyone involved. “Apart from EDA companies like Synopsys, there are private players who do offer training on flow and methodology,” says Synopsys’ Koteshwar. “The challenge they face is not the knowledge quotient for developing these trainings, but rather the availability of libraries, designs, tool licenses and exposure to different styles of design. Third-party training providers need to partner with EDA companies to offer comprehensive knowledge and skill-based training to design engineers.”

It can be difficult to separate languages from tools, even when they are industry standards. “I believe that engineers are best served by the vendor they are using,” says Siemens’ Willoughby. “For example, with SystemVerilog and UVM, we have the subject matter experts. Many of them used to be customers, used to be designers, and they have used the tools. They can go beyond high-level constructs and programming. An engineer is going to have to marry the topics they have learned with the tools that they need to do the job. There’s a role for contractors and independent trainers. But if I were to choose, I would prefer to get this knowledge and background from an industry expert. Then I know how to apply it to my job.”

It can be a tough call. “We are the subject matter experts, but they’re the ones who are experts on how to deliver that training, and how to deliver it efficiently,” adds Arteris’ Shuler. “They need to have people on staff who are capable of understanding it.”

Training is often an integral part of the business. “When the training is coming from a company that does things other than training programs, that actually sells software using the standard, then paying for the training would not be appropriate,” says Arteris’ Geday. “In addition, having the clients/students not paying for it gives us more freedom to use our software as an example. We do not have to systematically show all the products on the market, like the consortium would have to do if they were doing training themselves.”

There also are training holes that are only filled by the universities. “The industry needs to have more fundamentals type of training,” says Willoughby. “There is a lack of training for things like safety, or signal integrity. How do these issues fit into the whole scheme of things? Why do I have to worry about it? Then we can tie some of the tools into this discussion, because ‘that’s why you use this tool to be able to get this engineering concept done.’ Maybe we haven’t paid as much attention to it because the resources are always so limited, but it is a needed element. That’s an area where it would be nice to work with some consultant to come up with material like that.”

It is more difficult when the technologies and methodologies are still evolving. “We have done a lot of evangelizing about processor verification,” says Davidmann. “The methodologies are evolving, and so you can’t get a training course on processor verification. First, you have to get consensus on the way to do it. Then, you could try and formalize it, to communicate and educate people. Some areas never get to that point. For example, each of the major vendors has their own formal tools, and conceptually there are similarities, but they are all different. So there is no industry standard training because there’s no real consensus in the technology.”

Conclusion
Training is a cost and an opportunity, both for those who receive it and those that provide it. The industry has struggled to find an acceptable way to evangelize new technologies, and this has influenced some development. At other times, it appears that the EDA industry has done itself a disservice by attempting to create new languages when existing ones would have served, avoiding such high levels of resistance.

Perhaps most importantly, tools, languages and methodologies have little value if nobody knows how to use them effectively. “Training should be transformed into learning,” says Koteshwar. “That means organizations need to start looking at product training as more than a strategic lever, because it plays a larger role in the value we can provide to the customer.”

Related
Continuous Education For Engineers
Companies that invest in their employees’ education often get rewarded with more productive and happier workers.
Designing Chips In A ‘Lawless’ Industry
Mind-boggling number of options emerge, but which is best often isn’t clear.
Engineers: Job & Education Knowledge Center
Top stories, blogs and resources for engineering education and jobs hunting.
Stretching Engineers
The role of engineers is changing, and they need to be picking up new skills if they are to remain valuable team players. There are several directions they could go in.



2 comments

Theodore wilson says:

Thanks. These survey articles are invaluable service to the industry. Always thought provoking!

Ron Lavallee says:

Are evangelizing and training equivalent? Isn’t evangelizing about ‘getting others to try’, versus ‘getting others to use’? Tricky question that I am not sure has a correct YES or NO answer, but I am sure it’s possible to chart a few scenarios involving both, beginning with an idea. Most everything evolves and so does evangelizing and training. Evangelizing evolves to marketing and training evolves to support. It seems that teams or individual inventors who embrace this evolution succeed the most.

Leave a Reply


(Note: This name will be displayed publicly)