Re-shoring And Rebuilding The IC Supply Chain

Raj Jammy, chief technologist at MITRE Engenuity and executive director of the Semiconductor Alliance, sat down with Semiconductor Engineering to talk about changes in the supply chain, where and how to leverage different capabilities, and why advanced packaging and manufacturing are so critical to economic security.

SE: The global supply chain for semiconductors appears to be splintering. What’s the impact of that?

Jammy: The market is being reshaped because of the increasingly prominent role of the fabless-foundry industry model and the supplier community, whether that involves physical tools or software tools. Without these, there is no industry model as we know today. These are linchpin technology providers in the ecosystem. We are seeing foreign players excelling in manufacturing of some of these innovations that we have pioneered in the U.S., and becoming stronger in the marketplace. While that’s good between allied nations, we are working to regain strength and bolster our national economy and security. We need domestic manufacturing capabilities so that we’re not dependent on other countries for critical components. Economic competitiveness is good as long as there’s a level playing field. On-shoring chip manufacturing technologies is critical, and you also need packaging to go with it for a robust supply chain. End system makers need an assured supply of components. We are working with allied countries to make sure we have crucial technology capabilities, either onshore or with our allies, where we can depend on them and they can depend on us.

SE: What’s different now?

Jammy: It’s not practical to have the kind of mutually dependent relationships we have with our allies with everybody around the globe, at least in the current geopolitical situation. There are certain unique strengths and capabilities that some of our allies bring to the ecosystem, and we share common threads on many issues. Working together is always going to be mutually valuable. We also have to keep in mind that many companies headquartered in the U.S., as well as U.S.-branded companies, have facilities in nations that are allied with us. We should see how we can strengthen all of that.

SE: Where is MITRE in all of this?

Jammy: MITRE is a systems company that is focused on public good and security. We have nearly 10,000 people and 270 physical labs that work on a variety of topics of national and global importance across the technology spectrum to policy, to health care, and more. Semiconductors are in the middle of everything, and they are increasingly critical to national security and economic well-being. One of the areas we’re very good at is private-public partnerships. We’re also good at program management, and if you look at what we’re facing now in semiconductors, it’s a national problem. So we’ve formed the Semiconductor Alliance with a number of ecosystem entities in the country, where all these companies, universities, regional entities, and industry bodies come together on a regular basis to discuss and solve key challenges facing the industry. We’ve done this time and again in many other areas, and semiconductors is one area where we strongly believe we can make a difference. In the Semiconductor Alliance, everyone works together to find out what would be the best way forward to develop a vision for the National Semiconductor Technology Center (NSTC) and the National Packaging Manufacturing Program (NAPMP). What are the technical problems we should be solving through the NSTC? Should these solutions be tactical or strategic, or some blend of those? The solution set is ideally distributed across the nation, as each region has its own strengths. Our vision is that the NSTC is a distributed network, a whole-of-nation effort to solve a national problem.

SE: Three foundries are vying for process leadership, and there are additional companies partnering and competing with them in advanced packaging. We also have a slew of new technologies, such as backside power delivery, hybrid bonding, and potentially a commercial marketplace for chiplets. What does leadership actually mean in that context?

Jammy: Let’s take packaging as an example. For a long time, packaging was labor-intensive. It required a lot of manual processes, and a whole range of materials that needed to come in place. It had a natural migration to places where these conditions were easily met, and that was in Asia. But increasingly packaging requires automation, because the precision levels and the tolerances required for the kind of packaging we’re doing now, or which we’re contemplating doing, is becoming difficult to do manually. So the manual component is going away, and more automation is necessary. In addition, with packaging, there are lots of options in front of us. If we are able to standardize certain types of packaging options, the entire industry can focus and rapidly accelerate around those standards. That’s similar to what happened in semiconductors. We had the International Technology Roadmap for Semiconductors (ITRS), and it wasn’t that everything ITRS had was exactly the right solution, but because we all focused on a particular path, that helped us to get much more done. The same can happen with packaging. If we as an industry can define what this packaging is going to be, or how it is going to be best executed, then we have an opportunity to lead this industry again.

SE: So it’s no longer just a race to 10 angstroms or below? Is that now just one more factor to consider?

Jammy: The move toward finer-scale technologies is inevitable. You still have the performance gain, and you still have the power and area improvements. It’s more marginal, but the ability for the designers to have more devices and more circuits is always wanted. So people will continue to do that and take advantage of it — unless you provide another vector for them to do the same thing without scale. But we haven’t reached the spot where we can say, ‘I can give you the same performance without the power loss on a different die, or using some other method.’ Perhaps heterogeneous integration is one of those approaches where we actually can achieve some of that, but we have to think differently. It’s no longer a two-dimensional chess game. It’s a three-dimensional chess game that we need to get good at, and we haven’t yet shifted the paradigm. In the past, you had a chip and you would say, ‘Here is the latest-generation node. Go build something around it.’ System makers would design a laptop or a cell phone based on that. And occasionally you had these applications where people would advertise for ASIC designers because they wanted a custom chip for a specific application. But those were few and far between. The real powerhouses were the chipmakers that were trying to define the next-generation node. The industry has shifted away from that. Today it is systems companies saying what they want. If they can’t get it, they’re going to design it themselves and get it fabricated. It’s not that the core process capabilities and shrinks are not important. But it is using the base platform technology onto which you layer your design. That’s how you enable system-level thinking.

SE: That’s one of the fundamental shifts in the industry today, right?

Jammy: Yes, and it’s driving how we need to operate. We need to think differently. So how can we enable that paradigm to become more successful? It’s one of those areas we haven’t touched yet in semiconductors. There is a lot of opportunity for semiconductors to penetrate new areas and make our lives different. Five years ago, did we think what we have today would be possible? Probably not. The rate of change is accelerating.

SE: If the future is heterogeneous, and no longer just a single chip, how do you determine industry leadership?

Jammy: It’s by driving a full-stack approach. When we design a chip these days, it’s almost like designing a multi-component system like a smart phone. If you’re trying to put a phone together, you want to tightly integrate the hardware and the software to get desired functionality to make it more useful for the user. Similarly, when you have multiple chips with different functions in a package, it works like a single unit with functionality of many. If I’m changing a material — for example, a non-Si semiconductor for its unique functionality — I have to start thinking about what it does to my thermal budget and contamination of Si if it were on the same chip. Do I have to worry about any other issues with regard to the process, or migration, or any doping issues that might come up? What happens to the device architecture? How does this whole thing get designed? There are integration issues and packaging issues. You can’t assume it’s just going to be scaling and that everything else will fall into place. That’s no longer true. Instead, you can design a non-Si chip and manufacture it in a different fab at a suitable dimension and keep the Si chip at the best possible dimensional node and manufacturing process for performance and cost benefit. Subsequently, the two heterogenous chiplets can be integrated into a common package. As we do this, we are ensuring that everything from material selection to design to device architecture, integration, and packaging are all considered for the final end application and optimized. That’s what we’re calling a full-stack approach. And it means multiple players in the ecosystem have to work together.

SE: How do you ensure the supply chain will be able to provide all of the necessary pieces?

Jammy: The question is always going to be how many of those do you control? You should not be in a situation where you sourced a certain part but you can’t assure its supply in times of emergency. Or maybe you have a part that could fail catastrophically at a point that you don’t expect it to fail. Those are national security issues. And then comes the software overlay on top of the hardware component. You want to be able to manage these two, to a certain measurable and reasonable assurance level. It’s getting increasingly difficult to separate out national security and economic security. There is a physical component to national security, which may also depend on the source and provenance of these components. Then there is the economic security aspect, which could throw you into a downward spiral.

SE: It’s not just the latest nodes or most advanced packaging, right? There has been a huge gap between 200mm supply and demand.

Jammy: Yes, 200mm production has shown remarkable resilience. Let’s assume there is a certain 200mm-based component that is necessary to make the cell phones or automobiles we use. We can go and make those parts domestically, but it may take us a year to ramp up. So can we afford to completely halt manufacturing those cell phones or those autos? We need to ensure domestic or allied country capabilities for all components for better resiliency.

SE: There seems to be a global imbalance when it comes to materials — both natural and man-made.

Jammy: Materials are certainly one part. The other part is how to ensure the availability of these materials to make us a magnet for companies in the packaging space and the chip manufacturing space in order to attract them to come to the United States. But how and where we make all of these is extremely important. The supply chain is a very, very complex problem, and it’s not something that can be easily realigned overnight. There are so many materials, sophisticated components, and sub-components that go into making semiconductors today that it’s wrong to assume we can do everything domestically. While we can actually go and source many of these things, it is also important that we build some degree of balance and resilience into the system. We should build up a domestic supply source, but be able to secure components from an allied partner in case of any unforeseen disruptions. And we should back them up as well. That provides more strength for the whole fabric as opposed to having a single point of failure.

SE: Without a technology roadmap, where do we go next?

Jammy: There are thousands of smart minds trying to analyze different options. Is there a reason you need to scale, or can you just take an existing-generation node and do some tweaks to provide the performance you need? Perhaps you just need performance, and you can manage the power and area, so you can compromise between the different PPA paradigms. Scaling alone is taking us down a certain pathway, and there is a well-known advantage that comes with scaling that we are used to. So our first order of business has been, ‘Let me scale and then see where I can go with this. Is that giving me an advantage or not?’ People are increasingly realizing that we should start exercising the third dimension to see if we can gain that advantage without traditional area scaling, which is getting harder to do. At the end of the day, all the scaling is purely for giving designers more options in terms of how much functionality they can put in there without using too much power. Controlling power is going to be very critical in the future for enabling a number of new applications. You do need to keep the area tight — you have device parameters, circuit parameters, as well as the overall economics associated with that.

SE: And you need to add reliability, which is a fourth dimension, right?

Jammy: Yes, because as these devices get smaller and smaller, there are a lot of variances that kick in. With a 10-angstrom device you’re looking at atomic-scale spacing. So clearly controlling mother nature and atoms and electrons in that space is not something that is going to be so easy. While we do a very good job at it, we have to keep in mind that there will be variations that will kick in. If you look at the memory industry, they’ve done a phenomenal job in terms of providing error correction and ensuring they store information at a certain point in a certain block, and then they move to another block for data integrity. They’ve implemented this in a typical NAND flash chip that you can buy from a store. By doing this, they’ve increased the life of that particular package. There are similar things happening in other areas. How many times will a particular device go through computational cycles? And how many cycles can it endure before a failure potentially can happen? It’s not so difficult to do that. You just have to be able to provide that balance between what is scalable versus how much redundancy you need to provide.

SE: Given all of this, when it comes to security, we really need to secure the entire system, not just the chip, right?

Jammy: Right, security needs to be implemented at multiple stages. That’s not to say we’re forgetting the security that can be embedded into the chip itself. Security at the system level is one part of the program we’re doing. But we also need to look at supply chain security and resilience across the board. MITRE is very good at putting together public-private partnerships, and also very good at program-managing these activities to highlight the right kind of discussion points, such as security. One of the things we’re looking at with the NSTC, with our alliance partners, is how do we help achieve something bigger than what each entity or each part will be able to achieve? If there is a region in the country doing a lot of work on logic and another on memory, how can we solve problems that neither one of them could solve alone by working within their own construct? For instance, how do we do in-compute memory or near-memory computing? This potentially could reduce power requirements and accelerate processing without having to transport data back and forth. You can’t just solve these problems by assuming we’re going to have only a memory-focused activity or a logic-focused activity. Whether it’s design-technology co-optimization, analog, or whether it’s some future-generation terahertz devices, or even packaging that is necessary for all this, you have to leverage strengths of the individual players. But you also need to stitch them all into a bigger fabric.

SE: A lot of these designs are very customized, though. In some cases, the complex chips and packages under development are never even sold commercially. If you’re doing one-off designs, can you achieve the same kind of security as you can with billion-unit types of designs, where the weaknesses are well understood?

Jammy: That is a challenge. People will try to solve problems locally as much as they can. This is normal. And that’s typically, ‘I have a problem in my fab and I’m going to just go try to fix it the way I know it best. But sooner or later I recognize that the solution somebody else has may be much more suitable for a universal problem set, including mine. And it might be easier for me to adopt it.’ So if we can facilitate that kind of dialogue ahead of time, then people will make the right choices. That’s what we are hoping the nation can achieve through the NSTC and make us all more resilient and efficient. Our focus is providing the capability and program content that is going to bring everybody together to discuss and contribute. What are the specific technical problems that are necessary to be program managed and solved in the different centers and universities across the nation so that we can combine all these to achieve something bigger? It’s a lab-to-fab pipeline to commercialize ideas and concepts and accelerate innovation in the country in this vital sector.

Related Reading
U.S. Opens Funding Applications Under CHIPS Act
Opportunities open for production of leading-edge, current-generation, and mature-node semiconductors.
Where All The Semiconductor Investments Are Going
Details on more than $500B in new investments by nearly 50 companies; what’s behind the expansion frenzy, why now, and challenges ahead.