Ecosystem Changes

Experts at the table, part 2: What can go wrong and how to get things fixed when something goes wrong. Plus, the challenges of keeping tools and IP current when different parts of the supply chain change at different rates of speed.

popularity

Semiconductor Engineering sat down to discuss changes in the semiconductor ecosystem with Kelvin Low, senior director of foundry marketing at Samsung Semiconductor; John Costello, vice president of product planning at Altera; Randy Smith, vice president of marketing at Sonics, and Michiel Ligthart, president and COO of Verific. What follows are excerpts of that conversation. To view part one, click here.

SE: As the IoT takes root, everything gets connected across markets and technologies. How does that affect the overall supply chain and ecosystems within it?

Smith: The network is getting bigger. There are lots of relationships and partnerships. We don’t even know what to call them anymore because there are so many different kinds.

Low: From a foundry point of view, we see new products with creative ideas that translate to business opportunities for all of us. As these products are now connected, this impacts to overall supply chain including existing products. Comprehending the system requirements plays a critical role for us to prepare the right technology including ecosystem solutions.

Costello: The merger of the automotive and the IT industry is blending dissimilar expectations. In the consumer space, there is an expectation that you can update very quickly and it’s almost constant. In the auto space, if you have to go in to do an update of firmware it has to be scheduled and it’s a hassle. In a case like that, if people aren’t looking at how to blend those two they’re going to face challenges in the future. If everything is connected in your home, and something changes and your fridge and your lights and your washer don’t work anymore, that’s going to be a big problem. It’s bad enough if you can’t get access to things at work. But if you can’t get access at home—your kids can’t get on the Internet and the stove doesn’t work—that’s going to be a big problem.

SE: If you look at EDA and IP, they used to be one step removed from all of this. But as we go forward, everyone needs to work together. The partnerships have to evolve very fast. Can that work?

Smith: There isn’t anyone here who hasn’t talked with automotive companies about ISO 26262. It’s not something we thought about before. We’re trying to figure out with our customers who’s responsible for what. But we have to figure out how this gets done. Ultimately the end customer has to make sure everyone has done enough, so ultimately that partnership has to extend everywhere.

Costello: Standards will help. It’s impractical to have thousands of people get together to try to figure something out. The evolution of standards will become more important, both in the context of how things connect together as well as how do you really know that they work together. That’s one part that will help. The other part that is challenging is that when you’re putting systems together like this, someone has to be the system integrator and take responsibility for making sure the system works. They also need to have a functional model to show how other functions will connect into this. If that’s a standards-based approach, you can find a way where people can work together and have some kind of common interface. But it is important that someone take ownership, and in the automotive market it’s the carmaker. Their vehicle is the largest part of the value in the transaction with the customer and they have to make sure it’s put together correctly.

Ligthart: We’re far removed from the end users in the car. But standardization requires some experience. It’s amazing how many different interpretations of standards like SystemVerilog and VHDL are out there. We have very strong partnerships with Synopsys, Mentor Graphics and Cadence. We share our simulators and front ends. That has worked out really well.

SE: At the very front end of Moore’s Law, who’s collaborating on a chip?

Low: We collaborate with EDA and IP vendors and even our equipment partners. Transitioning to 450mm wafers is one example. How the tools are configured, how that works with software, we have to agree and ally with them. That alliance includes competitors, too. We are working with TSMC, IBM and Intel Corp. on that. It’s mutual. We understand we have to continue to drive Moore’s Law or there won’t be any more node migration. That’s for the bleeding edge. It’s the customer, EDA, IP, the design services partner. At times, we even get the end customer involved because we need to understand how the technology affects the end product and systems. Sometimes it’s technical, sometimes it’s supply-chain related. Everybody gets involved, but to what degree varies.

Costello: It used to be that as an FPGA supplier, we supplied a base platform and some basic EDA tools and IP, and that was all customers needed. That’s not the case anymore. We have much more complicated hard IP that we have to provide. We have an ecosystem of IP providers. We can’t provide a complete set of EDA tools. We have to rely on partners. We even have an ecosystem of design services partners. We have some customers engage with us early on so they can see the road map and figure out where the value proposition is. The other big difference is that in the past it didn’t really matter what our customers were doing with our products because we can’t really know who all of them are. Increasingly, we need to understand the vertical markets so we can anticipate customer needs. That’s a big change.

Smith: What’s changed is the nature of the customer. The architecture and the chip design used to be very tight. Now they may be in different companies and different locations. That changes things. We’re working early on with the architects. When they transition that to get the chip laid out, we provide information. The transition here is to service companies that may provide either one of those. We have all these different parties involved. If you go back 10 years ago, we were talking to an IDM. Now it’s spread all over the world.

Ligthart: We license to a bunch of different EDA companies and FPGA companies. What their end users do with that doesn’t make much difference to us. Our relationships are still one-to-one. But once the end user gets involved with a design they have simulated with X, Y, or Z, it parses fine, and now they use the EDA tool from someone else and it doesn’t parse. So then they come to us and ask, ‘Why doesn’t it parse?’ The answer is that it’s a violation of the LRM (language reference manual). Then we have to get into the conversation with Mentor, Synopsys or Cadence to figure it out.

SE: What goes wrong with these relationships?

Costello: One big challenge is the breadth of partnerships you sometimes need to engage with. Another challenge is one that involves backward compatibility or upgrading. The nature of the device or the tool or systems we work with increase exponentially in complexity. Trying to remain backward compatible all the time is a nightmare. As a commercial entity you have to consider at what point does it no longer make sense. Not only is expensive to do, but it also sometimes undermines the value proposition of what you’re providing in new tools or products. The customer is always right, but sometimes it may not be the best for the future.

SE: That’s a particular problem when some companies are looking at products with lifespans of 10 to 20 years, while others are looking at 1 year, right?

Costello: Yes, that is one of the pluses of minuses of being in the FPGA market. We’re not in too many consumer products, so long lifespans are engrained into this. And it’s difficult because our products may be radically different 10 years ago compared with what we’re going to be doing in two years time. It’s such a change in scale that it’s hard to maintain that level of compatibility. We do work to make sure we have 10 to 15 year product lifecycles, but it’s no longer automatic. If you want to be able to support a product that is 10 times more complex than what you did 5 years ago, you really need to look at upgrading the tools.

Low: This is true and typically linked to the market segment these companies serve in. Consumer products tend to have short product life cycles. Conversely products in the networking, datacenters and automotive space have a very long lifecycle. For the Foundry, it’s key that we understand our customers’ needs intimately so that we can prepare the runway for them whether is capacity or technology roadmap to satisfy their needs.

SE: Who do you blame if you don’t understand all the pieces and some of them are in development?

Smith: Often the party who figures out where the problem is or how to solve it isn’t the source of the problem. What plays into that is uneven behavior with respect to a sense of urgency. Different people have different senses of urgency in terms of how quickly that problem needs to get fixed. If there are personal relationships established ahead of time, it proceeds a lot faster. Our networks are growing, and the pain is often with someone you haven’t worked with before. You don’t know how to get them motivated, how to direct resources, what their priorities are.

Low: That’s because the trust bridge has not been set up yet.

Ligthart: There’s also a problem when someone finds an issue that isn’t critical. That’s when you start losing their trust. If you’re sending a critical defect every week and it’s because you pushed the wrong button, sooner or later you stop reacting to it.