IP Business Changing As Markets Shift

Semiconductor Engineering sat down to discuss IP protection, tracking and reuse with Srinath Anantharaman, CEO of ClioSoft; and Jeff Galloway, CTO of Silicon Creations; Marc Greenberg, group director of product marketing for Cadence‘s IP Group; and John Koeter, vice president of marketing for Synopsys‘ Solutions Group. What follows are excerpts of that conversation.

SE: What are the biggest problems in IP today?

Galloway: The most pressing issue right now is complexity. Systems, technology and customer requirements are getting more complex.

Anantharaman: The biggest issue we’re seeing is just keeping track of IP—who has licensed it, who is using it? If someone has created IP and a subsystem is using third-party IP, then if someone else uses it you need to track that dependency. The audit typically lands on the EDA manager’s desk in the form of Excel spreadsheets. Sometimes acquisitions cause problems, too. The headquarters of a newly acquired group doesn’t know where the IP came from. So the real issues is not knowing what’s there and being able to track it.

Greenberg: There’s a concern, but not necessarily a problem today, with IP protection. It comes down to good relationships with customers and knowing what they’re doing. A lot of the IP companies have good relationships with the customers. There may be instances where an acquisition took a tortuous route. But for the most part, it’s easier to handle by relationship than to add technology.

Anantharaman: It’s more than someone trying to cheat on this. It’s managing the complexity.

Galloway: Part of the problem is merging of groups and acquisitions. Some of these companies are so vast that IP purchasing and the groups using IP are disconnected.

Koeter: The biggest problem is that ‘build once, sell many times’ has largely fallen by the wayside. Every day a foundry has introduced new process technology. Our customers are working with 0.1 PDKs. The processes are churning. The standards are churning. PCI SIG announced PCIe 5.0. There is USB 3.1. There is DDR5, LPDDR5. Everything is changing. Schedules are getting compressed. Customers are consolidating to the point where they are becoming market makers. They want things their way. Maybe it’s trivial like a custom metal stack, or it may be significant because they look at this as a way to differentiate.

SE: What about subsystems?

Koeter: Clearly subsystems are happening, but they’re different than what we thought would happen several years ago. The whole ‘design one, sell many’ paradigm isn’t happening. So now we see customers trying to marry IP with their own SoC architectures. That requires a degree of customization. That’s a challenge because the SAM (served addressable market) is getting more fragmented. With that fragmentation it becomes harder to track IP usage. Inadvertent usage is more of a problem.

SE: The term we’re hearing for IoT is mass customization. How do you keep track of all of the IP that gets tweaked slightly?

Greenberg: We have fairly strong delivery systems in place. People know which version of the IP they’re getting.

SE: But does that stay on the next version of the chip, or does it get changed out?

Greenberg: There are different scenarios. Sometimes companies are producing a family of products. They may not want to tweak the IP from generation to generation, especially if it’s at the same process node. They architecture may be adding or taking away stuff, but for each individual piece of IP they will probably want to stick with the same version unless there’s a really major upgrade. But if you’re going across a process node, or if you’re going from DDR4 to DDR5, then you’ll take a whole new set of IP. The systems we have today are pretty sophisticated. I can’t think of any instance where someone got the wrong version of IP.

Koeter: From a development standpoint, the thing to guard against is that you can’t have segregation of your code streams. With mass customization, which is happening, you can’t fragment your code into 1,000 different versions. Otherwise, you might as well be a services company instead of an IP company.

Anantharaman: And with new versions of IP, you’d better deliver that to the CAD team or whoever is working with it. But the challenge is, ‘How do you let everyone who is using this IP know about this?’ If there is a new release, perhaps they need it and perhaps they don’t. You need to manage this like a dashboard. If anything changes or a critical bug is found, whether you’re using this directly or indirectly, it’s important to get that information out.

Greenberg: Yes, and as an IP vendor we deliver to a virtual loading dock.

SE: Is IP the same after you modify a system? And how do you track that?

Greenberg: If you’re taking up a new version, everyone on the design team needs to know what it is and why it’s happening. I would not advocate updates over the air. If you’re going to update a version, you’re going to do that for a reason. It’s not just because there’s a new version. You’re trying to correct a problem or add new functionality.

SE: But think about automotive. You want to make sure something is current for 10 years, and this technology is still being developed.

Galloway: Over-the-air updates would apply more to software. Code with RTL and hard IP, certainly not.

Koeter: We’ve been looking at OTA (over-the-air updates) very closely because there’s no question that OTA is the way of the future. You get updates on your end products. From an IP vendor’s standpoint, one of the things that’s critical to to make sure you have OTA. You have to have a trusted execution environment where you can get the OTA update, check for malware, validate it, make sure it’s authenticated correctly, and then you can execute. So what you’re going to see from an IP and a chip vendor perspective is a trusted execution environment. That’s critical for OTA.

SE: Doesn’t that change the way you interact with the IP? Today, most people have no idea what IP is in the products they buy.

Anantharaman: Most of our customers don’t even want to upgrade to the next software release. They are very conservative. They want to track exactly which releases of software were used. They will never let IP be automatically be updated.

Galloway: There is a hardware/software paradigm where the software is hungrier for updates and the hardware is more conservative. The automotive industry is very conservative.

Greenberg: You can always roll back software. But you can’t roll back hardware unless you have some programmable logic in there. Generally speaking, if there is no programmable logic there is no rolling back hardware after it has left the foundry.

Koeter: I’ve been hearing for the past 20 years about silicon limitations due to gate density. You can think about having a programmable accelerator like an embedded FPGA. There are a bunch of small startups in that area. You can see hardware that is largely programmable. But the reality is that foundries are adding new processes whose sole purpose is to save customers 3% to 5% die area. Memories don’t change. I/Os don’t change. Logic gets a little more dense. But customers are flocking to new processes to save that 3% to 5%. Silicon will never be free. So adding programmability into hardware is an issue.

SE: But concurrent with that, we’re seeing more fragmentation of markets, along with a bunch of new markets like autonomous vehicles or AI or the cloud, for which there is no basis of comparison. If a mass customization is applied to these markets, what does that mean for IP? Doesn’t that become more fragmented, as well?

Galloway: In some way, the two have to go together. We’ve seen a lot of process nodes introduced, and then they wither and die. The foundries want to offer more specialized processes that offer an advantage here and there, but the process comes out and the IP doesn’t exist. That makes it harder to make a lot of different chips. In some ways, the two have to go together.

Greenberg: Process complexity is a major challenge. The IP business model in the beginning was build once, use many time. Now, we’re dealing with customized features and lots of process nodes.

SE: But how do you choose? In May, Samsung introduced six new processes and a new packaging technology in a single day.

Greenberg: A lot of it comes down to experience in this industry—knowing who your customers are, what the foundries are doing, and which of the processes will survive.

Koeter: More and more of the choices are being made by the customers.

SE: But do you have enough lead time to make that decision?

Koeter: That’s the trick. If you’re doing a port from one process technology to another, six months or maybe even a year is not an unrealistic lead time to do a complex IP port. It’s a very dynamic situation. Our customers are keeping their options open. From our perspective, we can help mitigate that by engaging with the foundries. Are they willing to put their money behind the process? Sometimes the foundries will pay you to port to their process. But there’s an ongoing discussion. Nobody wants to build something that isn’t a win for the foundries and for the IP vendor.