IP Challenges Ahead

Part 2: For the IP industry to remain healthy it has to constantly innovate, but it’s getting harder.

popularity

The revenue from semiconductor IP has risen steadily to become the largest segment of the EDA industry. Industry forecasts expect it to keep growing at a CAGR of more than 10% for the next decade.

Part one of this article examined the possibility those forecasts are wrong and that large semiconductor companies are likely to start bringing IP development back in house. This is a minority viewpoint, and one not embraced by much of the semiconductor industry.

“The reason is simple,” argues Sanjive Agarwala, senior director of worldwide silicon development and fellow at Texas Instruments. “The number of nodes will slow down, industry consolidation, standardization of IP—these all add margin pressure. And R&D costs are proportional to what you can afford.”

Others are less pessimistic, with some qualifiers. Hugh Durdan, vice president of strategy and products for eSilicon adds another dimension to industry health. “One is revenue. Is it growing? It looks quite healthy today. The second dimension is profitability. That is where the real challenge lays for the IP business.”

The real question becomes whether the IP industry can transform itself to deal with the evolution of the market. As quickly as some companies may be re-aggregating IP development, other companies are entering the market. “We see major OEMs, who used to buy SoCs from merchant chip suppliers, now designing very complex SoCs in house,” says Tom Wong, director of marketing, design IP at Cadence. “This trend will continue as the number of companies capable of deploying “unicorn” SoCs continues to grow.”

A Unicorn SoC is a complex SoC that ships in excess of 100 million units.


Fig. 1: Global Semiconductor Market. Source International Business Strategies, Inc. (IBS)

IP business model
One of the challenges for the IP industry is the business model. When IP is developed in house, it is an engineering cost. When IP is licensed, there may be both an engineering cost, which is the license fee, plus a manufacturing cost for many pieces of IP, which is the royalty.

“There should be a success element that help two companies come together and to encourage each other to push the envelope of IP development,” says Grant Pierce, president and CEO at Sonics. “The royalty model may be too difficult a business model for every company to share their direct profit back to every IP company that contributed to the profit, but we hope to be rewarded based on the utility we provided in their design.”

Another problem for some IP companies is scale. “The market is such that IP vendors tend to be small companies and the customers tend to be big companies with lots of purchasing power,” remarks Dave Kelf, vice president of marketing for OneSpin Solutions. “If the big company has all of the buying power and there are multiple companies trying to produce the same IP, then it becomes a tough market and the big companies know that they can squeeze on royalties.”

“Part of the success story for ARM was that they were able to get royalties for their entire IP portfolio and that drives their margin and profitability,” explains Anush Mohandass, vice president of business development for NetSpeed Systems. “Successful IP companies are those that can drive this business model and get both a license component and a royalty component.”

So why do some companies not get royalty? “That goes back to the impact of the IP,” continues Mohandass. “You have to provide value to the designers of the SoC.”

Value can be provided in several ways. “First, we saw that customers wanted complete solutions, and that may include integrating a controller and PHY,” says eSilicon’s Durdan. “Second, things like the PHY for DDR are extremely challenging to implement and to get to timing closure. So another service is the hardening of the interface so that the customer does not have to deal with that. The next level is complete sub-system integration.”

We are also seeing customers looking for increasing levels customization. “An example is Adapt IP,” says Kelf. “While they are trying to sell the IP for royalties, they make more money from configuring the IP to fit into the chips. They charge a service fee to configure the IP and doing the synthesis runs.”

Another aspect to IP health is just making sure you do get paid for what you provide. “There is a lot of IP floating around that has not been paid for,” claims Warren Savage, general manager of the IP Division at Silvaco. “This is either intentional or not, but there is money left on the table. Eventually, with the help of IP fingerprinting, IP companies can protect their hard earned work and help semiconductor companies stay in compliance.”
Impact of Open Source
ARM is the shining star in the IP business because it is the largest provider of IP and it is quite profitable. Is RISC-V, an open source processor, a challenge to that?

“If people are invested with ARM or other cores, then having RISC-V be open source and potentially cheaper will not make difference to them,” says Geoff Tate, CEO of Flex Logix. “I don’t think people will be in a rush to put open source IP into their chips. IP has to be proven. They want to see timing and power numbers verified by actual silicon results. When mask costs are $5 million to $6 million and total chips costs are perhaps $100 million, they will not risk the chip to save some money. The important thing is to get it right first time.”

There are other barriers, as well. “It is all about the ecosystem,” says John Koeter, vice president of marketing for Synopsys’ Solutions Group. “We have a core – ARC, which is demonstrably better in a PPA compared to ARM. By every rational objective it should be sweeping ARM out of the market. We are talking about a 50% difference in area or performance. And while we are doing well, we are not a billion-dollar microprocessor IP vendor. It is all about the ecosystem and the software that has been ported to it and the development kits that are available.”

So, how does RISC-V stack up? “The scalable RISC-V instruction set means designers can choose the right mix of instructions, associated power footprint and throughput for the specific application,” says Graham Bell, vice president of marketing for Uniquify, Inc.. “Power footprints so far are competitive when compared to commercial offerings.”

While RISC-V may not directly attack ARM in its core markets, it may make inroads elsewhere. “One company had a controller in all of the SoCs,” explained Rick O’Connor, executive director of the RISC-V foundation. “This is not seen or programmed by the end user. It controls the SoC, and there are multiple instances of them on the larger SoCs. It was homegrown and they have been maintaining this for a long time. They needed to upgrade it to 64-bit. After they reviewed their options and considered internal development, they picked RISC-V. It didn’t steal any IP business, but it may have stolen some growth.”

Durdan agrees. “For high-end processors, such as those in a smart phone, there is a significant barrier to anyone else getting into that market. It is not insurmountable. But it would take a long time, mainly because of the comfort and familiarity that the software engineers have with the architecture and tools. But there is another class of processors that is more embedded. For those, all of the software is written in house and they do not rely as heavily on third party software ecosystem. In those applications it is much easier to switch architecture.”

Many see open source as a distraction. “USB and PCI express are effectively open source,” Koeter points out. “They are based on a public standard that anyone can implement. It is all about how well you can implement it.”

So RISC-V may take away some growth from CPU providers, but others argue this creates more opportunities. “It will bring more people into the fold,” says Mohandass. “People who are buying merchant silicon or buying standard components will start to do their own. It will reduce the cost of entry into designing your own hardware. It may reduce the IP revenue per chip. So will the increase in volume be enough to balance the decrease in perceived value? I don’t know the answer to that.”

O’Connor agrees. “It may lower the barrier to creating new devices. While they will not license a commercial processor core, they would still need to license all of the peripheral IP. There are many more of these than processor cores, so it may grow the market.”

In the meantime, the market watches carefully. “Linux certainly killed Solaris, but software is different from hardware,” points out Savage. “I’m not a big believer in the open source hardware movement for a variety of reasons, not the least being I don’t see customers asking for it. But regarding RISC-V, there are a lot of ways this could go off-the-rails on its journey to be the Linux of processors. I’m watching just like everyone else.”

Very hard IP
Another possible development for the IP industry may be enabled by advances in packaging. “Multi-chip packaging is a niche technology limited to a few specialized users in the past,” explains Tate. “We hear a lot more about it today. The willingness to consider a multi-chip design is going up sharply and people are considering it for a lot of reasons.”

“It is a reality today and there are people selling chiplets with certain functions embedded in them,” says Durdan. “The value to the end customer is that they can isolate a piece of functionality that may be high risk, such as a high performance SerDes, from the rest of the SoC. They are mitigating risk by using a known, proven chiplet that embodies the high risk IP.”

There are other advantages as well. “It makes the most sense to think about chiplets when you want to integrate things built on different process technologies,” says Drew Wingard, CTO at Sonics. “The image sensor may need low noise which you will not get from finFETs. Other things may be built using MEMs. There are key parts of the RF link that can be manufactured using Si or GaAs that don’t behave nicely with complex digital. There are good reasons to believe that heterogeneous integration at the package level makes sense.”

Typically 20% of an SoC design is analog or analog/mixed-signal. “This IP for SoCs will be developed at 2x and 1x nodes, or new ways will be used to combine older analog-friendly processes with current digital offerings through the use of 2.5D and 3D stacking technologies,” says Bell. “This can change the IP design problem to a packaging and manufacturing, one that can be easier to solve.”

But there are barriers that have to be overcome first. “The problem with chiplets for any IP supplier is the lack of interconnect standards,” says Tate. “You have to drive signals over an interposer. People use different types of interconnect technologies, such as HBM, and everyone seems to do something different for connecting the chips. That is the signaling layer and then there is the logical layer and there is a lack of standard here as well. We are not yet at the point where they can buy stuff off the shelf because of the lack of standardization.”

Cost is the biggest barrier today. “It is not going to be the lowest cost solution for the end product,” points out Durdan. “It is more expensive to package up multiple chips into a package compared to a large integrated monolithic chip. There may however be a time to market advantage to the chiplet solution.”

Those costs will probably come down over time leading IP developers to change. “Chiplets require silicon hardening,” says Bell. “Competitors will not find it easy to enter this market, particularly those that have a soft IP focus.”

Conclusion
Moore’s Law is irrelevant to IP,” says Savage. “Bigger chips mean more IP, and an increased number of smaller chips mean more IP. The IoT era is going to mean more IP licensing than we’ve seen with ASSP’s because instead of having one ASSP with one IP license, a company might have five small chips with five IP licenses.”

The IP industry had to change in the past, and it will continue to evolve in many ways. Semiconductor companies cannot afford to put IP companies out of business because their success relies on them. While each may want to take a larger part of the pie, at the end of the day they are business partners. IP companies that invest more in R&D will produce better products and will demand higher prices for it.

Related Stories
Is The IP Industry Healthy? (Part 1)
IP has grown to become the largest segment of EDA revenue, but is it sustainable?
IP Business Changing As Markets Shift
Experts at the Table, part 1: Rising complexity, new technologies, and the challenges of keeping track of and protecting intellectual property are straining the ecosystem.
Re-Using IP In Packaging
The biggest advantage in advanced packaging is still missing, but that’s expected to change.
Challenges Grow For IP Reuse
Methodologies for integration become a competitive tool as complexity and possible options skyrocket.
No More Easy IP Money (July 2016)
Revenues for semiconductor IP are expected to grow, but the easy money may have already been made. Developing IP is getting tougher on several fronts.