中文 English

The Chip Industry’s Next-Gen Roadmap

SRC’s new CEO sheds some light on next-gen projects involving everything from chiplets to hyperdimensional computing and mixed reality.

popularity

Todd Younkin, the new president and chief executive of the Semiconductor Research Corp. (SRC), sat down with Semiconductor Engineering to talk about engineering careers, R&D trends and what’s ahead for chip technologies over the next decade. What follows are excerpts of that conversation.

SE: As a U.S.-based chip consortium, what is SRC’s charter?

Younkin: The Semiconductor Research Corp., or SRC, is a consortium that was formed in the early 1980s. At that point, the drivers were largely 2D scaling, III-V devices, packaging, interconnects and architectural innovations. But at the heart, what we’ve always done is embrace academic researchers as a source of good ideas, and make sure they are grounded in interesting problems that, if solved, would enable and be adopted by industry.

SE: The SRC helps develop R&D projects, and then allocates funding for these projects, right?

Younkin: We primarily fund universities, with more than 130 contracted globally right now. The first step is the SRC works with industry and government to define a solicitation and secure financial support for a research program. So step one is to define the problem, kind of a DARPA-like model. We put the solicitation out into the open domain. We believe that competition is the best source of innovation and ideas. University faculty typically submit proposals in two stages, which allows us to work with industry and government to provide feedback that strengthens the initial concept. And unlike some funding models, where you leave the work alone, our philosophy is that when the project starts, that’s the time to get involved and make it better. That’s when you start asking the hard questions, or ask why something didn’t work and learn from it. And SRC shares that information with the project members.

SE: Is SRC funded by companies, the government and universities?

Younkin: In short, our primary source of funding is our 21 industrial members. The secondary source is government funding, where DARPA is our premium partner through their Electronics Resurgence Initiative or ERI. We’ve also partnered with NIST and NSF for many years. We do have some universities that provide cost sharing for selected projects. That’s a third and smaller pillar of our investments.

SE: The United States is a major innovator in the semiconductor industry. Yet there is a perception that the U.S. is lagging when it comes to funding basic research as compared to other nations. In general, what’s the state of semiconductor R&D in the U.S.?

Younkin: In general, the U.S. ecosystem is a bit bifurcated. We’re very good at both early-stage ideas and commercialization. But we’ve pulled away from the space between those two pursuits. I see that other nations aren’t afraid of that applied innovation space, or they don’t seem to shy away from the challenges that bridge the research and commercialization space. And for one reason or another, the incentive systems in the U.S. create a huge vacuum in that middle ground. Part of the reason I took the role as CEO of SRC is that I enjoy the middle ground and see SRC as a big element of success for our ecosystem. We’re working with faculty to shape ideas that are longer range and speculative, trying to distill them to practice and hand them off to industry. But we’re a modest-sized endeavor. While we don’t have anything near the budget of NSF (National Science Foundation) or DARPA, we have a great impact globally because we’re not afraid of that middle ground.

SE: China’s government is pouring $150 billion of funding to develop its semiconductor industry. The European Union (EU) recently launched a multi-billion-dollar semiconductor initiative. Does the U.S. government need to step up and provide more funding to the U.S. semiconductor industry to help keep up with other nations?

Younkin: This is a space where the government is the right answer. Our government is there to help identify and solve the big challenges that will benefit society. And investments in semiconductors as well as information and communication technologies are exactly that. Semiconductors are one of America’s top exports. They provide countless jobs and are a source of global leadership. But again, there is a bifurcated incentive system that separates industry and academics. The government can help incentivize and create energy around that. Talented people are expensive. It takes investments here. You also need to highlight and solve the right problems at the right time. So, that’s where the government can step in. I was pleased to see that the National Defense Authorization Act (NDAA) legislation recently passed by Congress. It is evidence of the increased appetite for semiconductor R&D that disseminates from the U.S. industry. Academics have been working with SIA and other channels to help government connect the dots between what society gets if we invest and what it loses if we don’t.

SE: What else needs to be done?

Younkin: We need to try to take a page from what China and the EU do well, which is collaborate where it makes sense and to not spread the dollars so thin that we make them ineffective and get nothing in return for the increased investments. What I would hope to see are some sensible near-term wins that take the ecosystem to the next level. What I would also hope to see are wise discussions about how people can partner and create something that’s bigger and better than the sum of the individual parts. SRC has a track record of excelling on the multi-party collaborative side of that equation.

SE: Still, China is outspending the U.S. and others in the semiconductor industry. How can the U.S. keep up with that?

Younkin: We still have advantages based on our collective experience and our shared interest in innovation. What we can’t expect to do is argue that we’re going to need or get China-level funding. That will not happen. But we can be smart about how we use our resources and the parts and pieces that we do have to gain and maintain a competitive advantage. Where I see that we will get into trouble is if we rest on our laurels, fail to innovate, and do not remain hungry. We must take these new dollars and say, ‘What can we do that really puts us ahead of the game in 10 years? And how can we do that together, so that we do it swiftly, learn from our mistakes as a group, and stay out in front.’

SE: Demand is increasing for engineers and related technical fields in the IC industry, but companies are struggling to find enough talent. Simply put, there is a talent shortage in the U.S. and elsewhere. How do we develop a new semiconductor workforce and how do we get more people interested in this industry?

Younkin: I started in 2001 as a chemist and material scientist. I was excited about shrinking things down to the chemical and atomic scale, and that was a compelling reason to go into the industry for material scientists, electrical engineers, and many computer engineers. In recent years, that pipeline has started to dry up significantly. To attract talent, we must share some of the hard problems that we don’t know how to solve, and try to illustrate why they’re exciting. We have to show bright people that there are fascinating opportunities for the next 20 to 30 years, the kind that you can build a career on as a scientist or engineer. We must try to engage them in solving these problems. For example, you have neuromorphic computing or brain-inspired computing, which is very promising yet complex. Quantum computing and photonics are very exciting frontiers that are making great strides. The younger student population is not as receptive to the 2D scaling forever story. That narrative is not finding a way into their hearts and minds. We must seek a new narrative that helps students understand that now is the probably the most exciting time to join the semiconductor industry. There is a bright future.

SE: The SRC and SIA recently released an interim 10-year plan called the “Decadal Plan for Semiconductors.” It’s a report that outlines chip research and funding priorities for the next decade, right?

Younkin: That’s right. Working with scientists from government, academia, and industry, SRC spent two years to develop the report that will serve as our guide toward 2030 and beyond. With SIA, we’ve called for $3.4 billion in additional federal R&D funding to drive developments in five seismic shifts — smart sensing, memory and storage, communications, security, and energy efficiency. The full report is at the publisher and will be released soon. In it, we hope the industry will identify their own R&D challenges or emerging challenges and join our discussion. We’re calling for a larger, collective commitment to the next wave of materials and hardware innovation where integration is critical to realize the full potential of AI, 5G+, and quantum computing that we all seek. This includes monolithic integration, heterogeneous integration, and elements that include 2.5D/3D packaging, plus techniques that we have not even started to dream up or research yet. We must commit to equipment innovation, materials, design, packaging, manufacturing, and workforce development as essential elements alongside the integration needs.

SE: What are some of the technologies and/or projects that the SRC is working on?

Younkin: One project that I see great promise in is our work at the intersection of logic and RF from a packaging standpoint. With investments in 5G, there have been amazing gains in the digital front-end for RF. We are working to put advanced RF elements into a package alongside logic, memory, FPGAs, and even GPUs. But there are significant thermal management and form factor challenges. Plus, there’s less expertise or know-how to tackle the intersection of these two traditionally separate domains. So our researchers are considering the SoC/SiP and design/architecture tradeoffs and driving toward both chip-driven and packaged solutions. Those investments are driven by partnerships with UC Santa Barbara and Notre Dame. Industry insight shared with academics is very helpful here.

SE: What about chiplets?

Younkin: We’re always looking at monolithic integration and 2.5D/3D packaging technologies. Others have been looking at it for a number of years. There will be many R&D opportunities in the assembly of disaggregated compute solutions in the years ahead. It is a fast-moving part of our industry.

SE: Today, chipmakers continue to march down the logic scaling roadmap with 5nm in production, and 3nm and beyond in R&D. What about logic scaling? How do you see that going?

Younkin: I put my faith in engineers and human innovation, and believe that the companies driving advanced logic will continue to do so for at least the next decade based on technologies I’ve seen. But each subsequent node requires a massive amount of human capital, and missteps are very costly. In parallel to that, the paradigm for value creation in chips is changing quickly. Companies must enhance their portfolio with SoC/SiP advances involving 3D monolithic integration, new memory types, packaging innovations, new accelerators, and soon the rise of novel architectures for next-gen AI. With those changes, they need to realize the performance gains they’ve traditionally achieved via 2D scaling. It’s very tough. At SRC, we’ve been focusing on the second part of this equation. We’re educating the workforce that companies will need to successfully pivot to new integration schemes, new packages and new architectures.

SE: What are some of the other technologies that the SRC is working on?

Younkin: One of the research areas that we’ve enjoyed seeing fast progress on, and is time relevant, is work we have been driving to accelerate genomic sequencing through new computer architectures or FPGA-based implementations. While much of it was targeted at genomics broadly, the work has been able to pivot quickly to help address challenges related to Covid-19. We have some good work out of UC San Diego that has been able to accelerate local testing and help the city. While improving genomic sequencing relative to state-of-the-art, the real opportunity is in providing personalized healthcare solutions. For example, if you have a form of cancer and I have that same form of cancer, the treatment that we might get would be extremely different based on our own sequencing and everything that the doctors can learn about our personal situation. That’s only going to be possible if we can bring the compute solutions down to a reasonable price point such that it becomes fast and accessible. So that’s a really interesting challenge that we’ve been working on the last few years and will continue to pursue.

SE: What else is the SRC doing in biotech?

Younkin: We have a program called the Semiconductor Synthetic Biology project. One aspect of that work is literally looking at neurons and asking how to characterize them as if they were devices that we know and love. What are their characteristics? What can we learn at the fusion of biology and electrical engineering? And what does that mean in terms of the medical technology or the medical device community? We’ve been exploring implantable devices and non-implanted scaffolds. So we’re increasingly seeing the fusion of bio and nano as an emerging opportunity for the community and for the entrance of new partners from medical technology.

SE: What else is out there on the horizon?

Younkin: We’ve made additional investments in Carnegie Mellon in augmented reality, virtual reality, and mixed reality. Researchers there have created the ARENA, an immersive cross-country virtual reality center, where they can bring in digital twins — something that’s in the real world and something in the virtual world — and have them synchronized in a platform. The information is shared in a meeting much like this across the country in real time. It’s a nascent research project, but is honestly the best tech I’ve seen since we’ve all started working remotely last March.

SE: I assume SRC and its partners are looking at AI, right?

Younkin: AI research and ideas are everywhere, especially at SRC. On the AI front, one of our researchers is Professor Tajana Rosing of UC San Diego, who has been pushing hyperdimensional computing. This is looking at very large arrays of data and trying to understand how to tackle them with several orders of magnitude of improvement. That’s a space that is generating a lot of keen interest. The architecture looks compelling. The speed enhancements are several orders of magnitude faster. But the question remains how can we distill this down to practice into a programmable working chip in the near term? I don’t think we have the answer yet. This is why we spend significant time driving shared conversations between the hardware and software architects. They need to talk more with each other to make meaningful holistic gains.

SE: What does this technology promise?

Younkin: Hyperdimensional, or HD, computing looks to tackle the information explosion facing us in the years ahead by emulating the power of the human brain in silicon. To do that, hyperdimensional computing employs much larger data sizes. Instead of 32- or 64-bit computing, an HD approach would have data containing 10,000 bits or more. But to fully realize its potential, our researchers must continue to develop new coding and decoding strategies, fast algorithms, and make a push into meaningful hardware demonstrations.

SE: Some of your efforts are bearing fruit and have led to new programs, right?

Younkin: Some of our work has grown, graduated, and taken on a new life. There was an SRC research program in our DARPA partnership led by the University of Illinois called SONIC. It looked to take lessons from Shannon-inspired communications and distilled them into computer chip designs that would operate in error-tolerant ways. Stochastic computing is another way it has been described. You’re not relying on a deterministic solution for any given device to always be right, but use systematic and random variation to design chips that behave more like the brain. The chips followed the 80/20 rule, doing the best they could with the information available and providing fast, accurate, and efficient computing solutions. As one element of that program, researchers developed a prototype of what is called a super-Nyquist density electrocardiogram, an amazing headset that was able to capture more information from the human visual cortex than a standard EEG. That research led to a $20 million DARPA grant at Carnegie Mellon as part of DARPA’s Next-Generation Nonsurgical Neurotechnology or N3 program.

SE: Wasn’t the N3 program aimed at develop bi-directional brain-machine interfaces for service members? According to DARPA, these interfaces would be used to control unmanned aerial vehicles. Or they could be used to team up with computer systems to perform various tasks.

Younkin: Yes. I’ll stick to a commercial scenario. In the future, I can see us wearing a comfortable and stylish headset that reads our brain signals, and then feeds the data to our heads-up augmented reality display. Here’s one example. Let’s say a car is coming in on the right. It’s an electric car so you can’t hear it. So the system senses and notifies us long before our eyes see the car, and we avert disaster. As humans embrace more technology, we’re going to find that we’re augmenting our body with watches, earphones, head displays and implants. We are looking for the easy solutions that don’t require a lot of human involvement.

SE: SRC and your partners have formed the Center for Converged TeraHertz Communications and Sensing. What’s that about?

Younkin: We’ve been driving higher frequency devices and systems that look beyond 5G frequencies at the millimeter wave and beyond. We are trying to drive what we believe could be a 6G infrastructure, looking largely 140 GHz and up to terahertz-based chip designs, integrations, and system implications for handsets and RF base stations. Second, elements of the same hardware and designs could also be used for imaging and sensing. Think of it as a next-gen lidar kind of solution. As mentioned earlier, we’re putting added emphasis on RF and digital packaging integration at the intersection of what has been communications and logic. These are two traditionally separate buckets.

SE: What is on your wish list at the SRC?

Younkin: I’d like to see us get some funding for added investments in quantum computing and photonics to complement our current and growing research portfolio. The U.S. government has recently made some major investments in quantum computing. That was great to see.

Related Stories

Regaining The Edge In U.S. Chip Manufacturing

What’s Next In AI, Chips And Masks

Going On the Edge



1 comments

Gil Russell says:

Mark,

Nice to see you’re finally able to mention hyperdimensional computing in print. Suggest you look into Mohsen Imani’s work at UC Irvine.
Imani received his PhD from UC San Diego (Tajana Rosing’s Group).

Leave a Reply


(Note: This name will be displayed publicly)