Strengthening The Global Semi Supply Chain

Within the semiconductor ecosystem, there are a number of dynamics pointing to the need for new ways of partnering in more meaningful ways that bring resiliency to the global semiconductor supply chain. One of these is the move to bespoke silicon, stemming from a shift in the companies that create most SoCs today — the hyperscalar cloud providers. These market leaders know their workloads so well that they now create their own highly-tailored chips.

Compounding that, the end customer base is shifting from traditional chipmakers designing for a socket to a combination of systems companies, existing chipmakers, and a slew of new players that are developing chips for very specific applications. Some of those are being developed at the most advanced nodes, particularly for data centers and AI applications.

This confluence of factors is creating disruptions around the globe.

“Today, the bar for creating innovative bespoke silicon designs has been raised to over $50 million per chip,” said Vic Kulkarni, chief strategy officer at Si2. “The associated costs of IP qualification, architecture, physical design, creating a software stack, prototyping, and validation, are soaring from $400M for N5 to over $600M for N2 to make it successful. This has resulted in an elite class of IP, SoC, and silicon-to system designers.”

These soaring costs are the result of complexity of the silicon-to-system design flow.

“What we need is the best-of-class, multi-vendor, multi-cloud and hybrid cloud infrastructure, to achieve innovation, and at the same time meet time-to-results (TTR) and time-to-market (TTM) goals. And given the complexity and interconnectedness of the industry, no one company can know or do it all,” Kulkarni said.

Fig. 1: Near-future cloud infrastructure vision. Source: Si2

Others agree. “The increasing product complexity of bespoke silicon requires an ecosystem of multi-vendor workflows,” said Corey Mathis, director of software partnerships and corporate development at Keysight. “While some companies may indicate their platforms allow for end-to-end design, simulation, and verification, no single company is the market leader in every facet of the workflow. For semiconductor companies to get their complex designs to market in a timeframe that meets their customer’s needs, it requires best-in-class tools from multiple vendors.”

It also requires IP, equipment, packaging, and other components, which may be sourced from multiple regions.

“Today, there’s an awful lot of angst around everyone wanting to be self-sufficient in electronics and semiconductors,” said Tom Beckley, senior vice president for the Custom IC & PCB Group at Cadence. “But that angst has been there throughout the 40 years I’ve been in this industry. I can remember when Japan was a powerhouse and people were upset. Korea became a powerhouse and people became upset. Taiwan, too, of course. Then, TSMC created an incredible business model on the fabless side. There’s a lot of talk right now about how to regionally, or maybe through partnerships, have an independent and self-sustaining ecosystem across all those pieces. EDA, IP, and chiplets are all part of that, and groups have to work together. We can’t dismantle or re-create what’s been created very efficiently and very economically.”

Re-creating a supply chain is a gargantuan challenge. Beckley pointed to a study by Boston Consulting Group, which estimated the move to regional supply chains would cost $1 trillion in up-front investment per region, or per group of partners, and the cost of the underlying components would probably go up by 50% to 65%.

“That’s a really expensive way to go,” he said. “The opportunity is for us to figure out how to encourage the current ecosystem to work together more effectively, more efficiently, and to cross-pollinate that. To that end, we’re watching TSMC and Samsung go beyond Asia, going into the U.S., Japan, and Europe and building fabs. That type of direct foreign investment should be encouraged. At the same time, highly skilled labor should be free to move and move effectively without any restrictions on it. All those pieces together create tremendous opportunities for all of us.”

Current geopolitical tensions make that more difficult. “One of the key things the last three years have taught us is that the supply is critical,” said Rahul Goyal, vice president and general manager of product and design ecosystem enablement at Intel. “Even a small semiconductor part can put an entire manufacturing line on pause. That is a problem. We need a more resilient supply chain, a more balanced supply in the world. That’s where Intel is investing. We’re investing in the U.S. and the Western Hemisphere with the intent to make the supply chain stronger in the Western Hemisphere, and to balance the supply across the world. From an ecosystem perspective, the complexity of what Intel is doing — whether it’s products, or whether it’s the entire supply chain for the product lifecycle — requires partnerships across the board. Even if you’re not going to foundry, you still need to partner with EDA because you have to get the EDA tools ready for your design teams to use. The foundry takes it to the next level to make sure that the partners are a diverse set of industry players, because the customers will decide what they use. We won’t always know what tools they use or what IPs they want. So we need to get ready for all of that.”

The ecosystem will become increasingly important for this, Goyal said. “The industry is collectively trying to achieve an ecosystem in which we can all feel secure, we can have data transmitted the right way, with data ownership in the right places. Connected to this, there’s a lot of talk about analysis and security standards to make sure that we maintain the security of the data as we go forward. There are some challenges, but also there’s a lot of opportunity as we look forward.”

Part of achieving a more robust ecosystem includes embracing the cloud, noted Michael Buehler-Garcia, vice president of product management at Siemens Digital Industries Software. “The cloud is going to run the same model as a semiconductor foundry. And if everything has been designed in a cloud, that part of the ecosystem could be anywhere. This means the designers come to the data, as opposed to data and the tools going to the designer. But it has to have the right business model and the customers. We as an EDA ecosystem have to be ready for the technology, and for the customers to decide just like we did when we made the move from IDM to foundry. With that there’s all kinds of opportunities.”

And those opportunities extend to all parts of the supply chain, particularly EDA and IP. “There are a number of roles for EDA and IP companies to address opportunities to help make the ecosystem be successful,” said Bari Biswas, senior vice president in Synopsys‘ Silicon Realization Group. “We are moving toward toward systems of chips, instead of just a system-on-a-chip. We will see multi-die systems in next 5 to 10 years as we go toward this trillion-dollar market, and these are opportunities that we need to work together to solve.”

Biswas said power, performance and area constraints are still driving design optimizations, but he noted that security should be a fourth vector. “If the cloud could be a foundry house, then we need to bring security to the hardware and create silicon that is secure by design.”

So what is the best way to create a robust EDA and IP ecosystem in order to strengthen the global semiconductor supply chain? It’s complicated, but in the end what is required is good alignment between management from both EDA and foundries to enable workflows that meet customer needs, Keysight’s Mathis said. The current work between Keysight, Synopsys, and TSMC, is a good example of this.

Building a better supply chain
Nevertheless, the role of partnerships in creating a really strong global semiconductor ecosystem is not always straightforward.

Mike Eftimakis, vice president of strategy and ecosystem at Codasip, pointed to a paper on the topic by researchers at the University of the District of Columbia and Georgia Tech, which analyzed the impact of collaboration on the global relationships and global supply chain. “The researchers were trying to explain what the effects are, and how the relationships ultimately impact the performance of the companies and the supply chain in general,” he said. “They proved that the path is quite complex. It’s doesn’t magically work. It’s a chain. It goes from starting with a collaboration project, where that project or the fact that you work together improves the relationships between the companies that work together. This in turn provides operational benefits like, as they work together, they start to be better at what they do. That’s the operational benefit that directly impacts the performance of the companies. It’s not directly working together that provides benefits. You have this chain of events.”

Eftimakis said this is evident in the chip industry. “We start to work together, but until we really have focused on trying to gain results that will benefit both companies, there’s no real benefit in the end. So partnerships work only when the two companies start to do meaningful work together. Sometimes we see partnerships that are just marketing partnerships. They have absolutely no value. In fact, there are partnerships announced solely for marketing purposes just to make noise, but everybody understands very quickly that it’s not really a partnership if it doesn’t provide value. And if it doesn’t provide value, it doesn’t provide value to the customers, either. It’s really important to think deeply about partnerships in the sense that they need to actually build something. What can we do with another company to truly create value together? Not just, ‘We need this. Let’s talk to this company and we’ll get it.’ It’s more about what can be made together to make sure that we gain, and they gain as well, from the work that we did together. This approach has been shown to positively impact the global supply chain.”

Arm, for one, successfully leveraged a broad ecosystem to gain traction in a number of markets, from smart phones and IoT to servers and PCs.

“One of Arm’s legacies over our 30 years is we don’t call our customers ‘customers,’” said Chris Bergey, senior vice president and general manager, Infrastructure Line of Business at Arm. “We call them partners because it runs very deep with us relative to partnering up with other leaders in the industry to help from an enablement point of view. From a supply chain point of view, Arm works with all of the major foundries, and is also working closely with many of the third-party EDA companies. Arm is agnostic relative to what types of solutions our customers want to build and who some of their preferred suppliers are. Related to this, one of the things that makes Arm so unique is when talking about the heterogeneous data center, what that really is, is CPUs with some custom logic or specialized processing. And if you look at the Arm ecosystem of semiconductor companies, it is the who’s who of companies that build DSPs, the companies that are the greatest at I/O integration, the companies that are tremendous in networking spaces, etc. So, it’s really about that partnership of being able to take from Arm computing with a very performant fabric, match that up with complimentary IP from EDA vendors or other IP providers, and then build on top of that and add some sort of special value add that can be very powerful from a power efficiency point of view, from a computation point of view. Lastly, it can be built in multiple fabs, which provides flexibility. We are only able to provide such a diverse set of solutions through our partnerships.”

Arteris IP, for example, is working with Arm in the automotive space to speed up the decision process. It recently announced an expanded partnership with Arm said he expects to be able to help customers converge much faster on their next wave of designs. “A lot of people are taking electrification much more seriously, and the underlying global climate challenges are driving even more discussions,” said Michal Siwinski, chief marketing officer at Arteris IP. “Some of the European and Chinese OEMs already have declared that, starting in 2030 cars, will only be electric. No more sales of gas cars. California is trying to make the point that in 7 years, there will be no more sales of new gas cars. We’re almost in 2023 already, and considering the classical automotive cycle design cycle is 7 years, there isn’t that much time to figure out all of the requirements and start innovating much faster. Both Arm and Arteris already see this extra urgency. Even though it looks like it’s many years from now, we’re almost late already in defining the architectures and how all of these pieces need to be designed. The push on core innovation is through the roof. If we can do this better together, we can save a lot of the back-and-forth iteration, or potentially misalignment of customer needs.”

One size does not fit all
This is not simple, however. Keysight’s Mathis pointed out that that different companies segment partnerships in different ways. “They could be technology partners that enable cloud computing, foundry partners that provide crucial process development kits (PDKs), or workflow partners that integrate various tools for design, simulation, and verification.”

Si2’s Kulkarni noted that barriers to entry must be lowered from both financial and technology investment points of view for a larger ecosystem. “The next wave of innovation will come from a system-first mindset for the silicon-to-system innovation with ecosystem collaboration,” he said. “During a recent Si2 roundtable discussion, the complexity in a 5G to 6G realization was eloquently described by Dr. Mallik Tatipamula, chief technology officer at Ericsson Silicon Valley. He shared an important of vision of a multi-dimensional ecosystem mindset to make the transition from 5G to 6G a realizable goal for the silicon-to-system collaborative within next five years.”

Fig. 2: The emerging collaboration model which will create the big winners. Source: Si2/Ericsson

Conclusion
But while models can be created, Kulkarni added that some problems that can’t be solved via a simple partnership. “Either there have to be many players onboard, causing logistical problems and antitrust concerns, or, if the final goal requires an infrastructure that does not exist, or exists only in proprietary form that no else will support, then you must consider an R&D Joint Venture like Si2. We have been successfully using the collaboration model of members from IP, chipmakers, EDA suppliers and silicon foundries. Together, these members will survey the industry to uncover gaps in technology interoperability in the context of a complete workflow, create a laser focus on the key deliverables for the collaboration team, prepare high-value use cases and roadmaps, and then, together, work on a proof-of-concept reference implementation leading to an industry standard. As we look forward, especially in the era of chiplets, and bespoke silicon, the new considerations will be factory automation data, multi-die assembly rules and standardization, 2.5D/3D IC configurations, impact on test yield and so on.”

Finally, another new factor to consider is that the walls between electrical design, mechanical design, thermal analysis are starting to come down and blend together. “The need for EDA to be more interoperable with multi-physics CAE workflows is becoming more important,” said Keysight’s Mathis. “Even if there are areas where two software providers compete, there are likely more areas where their tools and technologies are complementary, so comfort with co-opetition is becoming more common.”