Changing The IP Supplier Paradigm

Just a few years ago, the (IP) business consisted of small blocks being sold by small companies and an almost over the wall delivery mechanism. The industry quickly realized the problems with this supply chain and the IP business went through very rapid change. At the same time, the average size of the IP blocks has increased and today, what we think of as IP, used to be the whole chip when the IP market was first born. We are beginning to see more changes in the IP supply chain.

Semiconductor Engineering sat down with Rich Wawrzyniak, senior market analyst for ASIC and SoC at Semico Research; John Koeter, vice president of marketing for the Solutions Group at Synopsys; Mike Gianfagna, vice president of marketing for eSilicon; Peter McGuinness, director of technology marketing at Imagination Technologies; and Michael Mostovoy, senior director for ASIC development at SanDisk Corporation. What follows are excerpts from that conversation. Certain members of the panel also wanted to point out that the comments belong to them and may not represent the views of the company they work for.

Semiconductor Engineering: What is changing in the IP supply chain and what impact will it have on IP companies?

Gianfagna: The usage of IP has become very prevalent. Today 70% of the chip may be reused IP, or in some cases only 15% of the chip is original work. More than half of it is memory. But the real story is not just the prevalent use of IP, but how you pick the IP. You have got all kinds of technology choices, technology flavors, vendors, interface standards. How do you deal with that? I think that the new battlefield is an efficient and transparent way to find the right IP and that the configuration of building blocks is going to hit the PPA for your chip. It is really hard to know that upfront.

Koeter: When I talk to our customer base, I consistently hear they are spending more than 50% of their time on software versus hardware. This includes firmware, drivers, boot code, diagnostics, RTOS and typically some middleware, such as a TCP/IP stack. For system performance reasons, they have to make sure they have the right configuration for the IP. Even if you consider something as simple as USB, there are many things that you can do in the configuration to achieve a higher sustained bandwidth or throughput. Customers require more surrounding the IP titles. Maybe it’s drivers, maybe it is help prototyping, maybe it is help customizing and integrating the IP.

Wawrzyniak: One of the ways in which customers select an IP provider is by the amount of value add they provide. By this we mean the components that help the customer use the IP more fully and productively. The less effort they have to put in, the more effort they can put into architecting and building their solution. The smaller guys may have IP that is very good, but if the customer has to work harder to implement it, it may provide less value than a piece of IP that may not be the greatest but has lower overall development costs.

McGuinness: That is a key point. The whole point of developing IP and offering it to another company is to solve a problem that they have. The problem is not necessarily the actual technology or IP, but it is bringing it together and integrating it into a chip. By virtue of the kind of IP we develop, we are very familiar with the hardware/software co-development model. With graphics there is a huge driver staff. The newest area of IP we are getting into is the camera and that starts off with an expensive software stack. All of the problems associated with integrating that – the software platform is as expensive, if not more expensive, than the hardware. If there is a shift that we see, it is that we have to move further up in the food chain. Instead of providing components, we have to provide much larger assemblies of components, integrated as sub-systems, because they can get huge efficiencies by doing that.

Mostovoy: IP integration into an SoC is very important and we need to do both hardware integration and system-level design. We need to choose our IP supplier carefully. IP needs to be chosen as early as possible and needs to be validated in the system and in the environment including hardware integration. How do you ensure the IP you are buying will help you and provide the right level of performance? When you integrate it with your SoC, will it work from both a functionality and a performance point of view? Many companies are facing challenges in this respect.

SE: The trend is clearly for larger blocks, often with differentiation in software. Is there a point at which we could conceive of a fabless, IP less design company? Can you create viable chips using only IP from other companies? Is the selection and integration of IP enough of a differentiator?

Gianfagna: I don’t think so. The industry doesn’t like to be homogenized and this would represent a leveling of the playing field. It is appealing at one level where you had reference hardware and you could build software on top of that. The problem is that nobody will be satisfied. You can differentiate more in hardware, you can make it faster, you can implement a new standard more efficiently in hardware than software, and there will always be a drive to differentiate a product at the hardware level. Another practical matter is that these are very complex chips and if you are just making a few of them, the economics collapses.

Koeter: Can you just have a sea of processors and just differentiate in software?

McGuinness: It has been happening for years within Imagination. We create chips for other companies and they have no knowledge about the graphics domain at all. There was one company that wanted to create a digital radio but they had no digital design experience. We built them a SoC for that. They had the RF knowledge, but the application processor and all of the software was written by us. They are a fabless and IP-less semiconductor company. There are several more that we have been involved in. We see many people who say to us, ‘Our value add is in system integration and the logistical management of supplying OEMs with this chip, but we don’t want to design and build the chips.’

Koeter: I am skeptical because I believe that everyone will want to differentiate in hardware.

McGuinness: But we do differentiate in hardware. We do a custom chip for them.

Koeter: If we think about Project Ara, it is a smart phone that is modular and based on standardized specs that you can plug things together. Facebook has been saying they don’t want boutique servers. Instead, they want standard servers that will be interoperable. So there are some top-down pressures that are starting to drive people in that direction, but this will not be the normal case.

Mostovoy: I believe there is another driver for hardware differentiation – power consumption. This is becoming more critical, and as you implement more things in hardware you save power. Even if you use IP from somebody else, you need to have enough competence to understand how the IP works, how it is going to be integrated, and be able to work with the vendor for both integration and possibly some modification of the IP that would allow you to differentiate.

Wawrzyniak: In the case of Imagination, they are an IP company and they have access to IP. If you are an IP-less company you have to concentrate on the integration side of things. If they are going to license all of the IP, it comes down to how they are going to license the IP. Do they have to license for each chip individually, or do they have a plan for a series of products and can negotiate a multi-use license? With single use model, they are not going to be competitive. Their costs will be too high and much higher than someone with a library of parts that they have access to. If the IP is not the major cost, then it may be different.

SE: Gary Smith said that IP reduces the cost of the design by 44%? If that is the case, then isn’t buying all the IP the cheapest way to do a design?

Wawrzyniak: If you can get access to the right IP, at the right cost – yes.

McGuinness: If you are confident that your value-add lies elsewhere, then you should be OK. There is always value in knowledge of the system. If you don’t put the system together yourself, then you are at a disadvantage. If you are dealing with an IP company that can’t provide this either, then you have a big problem.

Wawrzyniak: The idea of approaching this from a system-design standpoint is a great idea. The concept of IP sub-systems is that they are system-level functions not blocks. It has moved up one layer of abstraction. If you are dealing with a design at that level, the system integration task becomes a little easier. You are no longer down in the weeds.

McGuinness: And you are allowed to take various components into account when you do the system-design which changes the problem, or gives you the opportunity to innovate.

Wawrzyniak: The challenge for an IP company is to understand what their customers want. If you are moving from USB 2.0 to USB 3.0, can you just remove one block and replace it with the other? It is pretty easy to say that, but people would agree that this is an attractive concept. The trick is to architect the sub-system to make that possible. That puts an increased burden on the IP supplier.

Mostovoy: Companies always face a TTM (time-to-market) pressure. This type of approach can save TTM, but you lose your value add for the architectural perspective, from power reduction, performance. If you build a prototyping system early enough, then you can perform firmware and hardware co-development, you can achieve the same TTM advantages without compromising on the architecture, power consumption etc.

Wawrzyniak: You should be able to build this into your design flow without making those compromises. This requires a fundamental change in the way they approach the design process and they are not going to do it unless they see TTM advantages, reduction in cost, reduction in power etc. You won’t do it unless you believe that these are possible up front.

Gianfagna: If we were having this conversation perhaps eight years ago, we would have been discussing the chip business and we would have said that the chip industry is getting harder, fabless companies are losing value and they have to have software, a reference platform and more. Fast-forward and we are saying the same today except it is now about IP blocks. The IP block is the chip company from eight years ago. Everything is just happening at larger and larger size. What used to be a board design problem is now a SoC problem, where each sub-system used to be a complicated chip.

Part two of this roundtable can be found here.
Part three of this roundtable can be found here.