It’s about 15 years the concept of IP development and its usage took place. In the recent past the semiconductor industry witnessed start of a large number of IP companies across the globe. However, according to Gary Smith’s presentation before the start of 52[SUP]nd[/SUP] DAC, IP business is expected to remain stagnant for next 5 years. There are reasons to believe into Gary’s thesis. A bird’s eye view shows an IP sitting at the heart of an SoC or subsystem. This is significant reason for a system company to assess an IP fully before utilizing it; also assess the IP provider’s quality and other business practices. At the tip of the iceberg it appears very simple to buy an IP and use it in your SoC design as required. However there are significant implications of using an IP from business as well as technical perspective; not all system companies have bought into the idea of using 3[SUP]rd[/SUP] party IP, barring some standard and common IP blocks from reputed suppliers. The standardization of IP blocks that go into most of the SoCs reduces cost for the overall value-chain of developing SoCs; however it can commoditize the stuff to an extent that it can start impacting the differentiated value of SoCs. Moreover, there are serious technical implications that need to be considered before using IP. There has been a significant change in the modern SoC ecosystem where the system companies are experiencing increasing need of customizing IP before their use in SoCs.
Considering it from a macroeconomic angle in a consolidating semiconductor industry, the IP-based business model of SoC design does provide a good proposition provided differentiated value is added into the SoC. However, it’s essential that the hidden costs in accomplishing some specific tasks to make this model successful are better understood. Often certain tasks are not performed adequately because of lack of understanding, and also because the associate costs not considered. This can leave an IP in a poor state, inside or outside of an SoC. The success or failure of an IP in a system depends upon how best these tasks are understood, invested-in, and performed by the IP provider as well as the system integrator. There are specific costs involved in doing these tasks which stand apart from the usual developmental cost involved in the normal course of IP development. These exclusive types of costs other than normal development are mentioned below along with what incurs those costs and their proper rationale.
Cost of Differentiation – The differentiation in an IP has to be construed from the design level. The system companies are expecting differentiation in IP that fits into their designs so that they don’t have to design the same IP themselves as much as possible. A common form of differentiation can emanate from IP vendors for providing extended solutions such as interconnect along with the cores. It’s true that such differentiation can again be seen as common in an IP for different SoC vendors; however it can move the IP to a level up. IP providers such as ARM, Synopsys, Cadence and some others are providing subsystem level IP solutions. On the other side of the coin an IP provider can work in joint collaboration with an SoC vendor to design a completely differentiated IP. In this case the cost can be very high; in-sourcing of the complete IP team may be preferable for the SoC vendor. In other words, the IP team in-house with the SoC vendor can work at the sub-system level which allows the team to add enough differentiation, do trial layouts and optimize, and thus reduce risk and time-to-market.
Power is becoming a prime criterion for differentiation, especially in the mobile and IoT market. An IP characterized for certain power parameters with a particular technology needs to be re-designed in most cases of newer technology; moreover the dynamic power profile may change significantly with the use-cases.
Considering PPA (Power, Performance, and Area), an IP can be designed to have flexibility to scale between different factors such as power and performance according to the technology used.
A new concept of IP abstraction is coming up where an IP is delivered at a higher level of abstraction which goes through High Level Synthesis at the SoC end. This provides scope of differentiating the IP and SoC in power consumption. Qualcommand Googlehave used this approach with Calypto(now Mentor) HLS solution where they deliver ‘C’ code which can be further optimized while integrating into SoC at system-level. A start-up Adapt IP provides option for delivery of IP at a high level of abstraction. In this scenario, the cost for IP vendor can decrease, but that gets added up in the SoC for the SoC vendor to account for differentiation and implementation. Moreover, this brings in a newer methodology for SoC architecture exploration and integration at the system-level and asks for fresh investment and learning. I will talk more about it in later sections.
Cost of Customization – One may think customization as a part of differentiation, but actually they are distinct. Customization is a process which takes place during integration of an IP in an SoC. There is a separate section on integration in this article. In this section I am talking about the provisions which need to be made in the IP itself to make it customizable according to different environments; interconnect IP is a good example in this case. For example, ArterisFlexNoC can be configured and customized for interconnections on the chip that can provide best latency, least congestion, and optimize other aspects. Similarly power management can be another area where configuration can be added for power harvesting. So the question is how configurable your IP is so that it can be customized according to different environments? Configurability in your IP adds provisions in your IP to be customized in different environments. This increases the value of your IP to operate in a wider range of possibilities. More often than not an SoC vendor may need to ask an IP vendor to add specific customization in the IP so that it can exactly fit into the scheme of the SoC. This situation may get extrapolated to an extent that the IP gets transformed into a subsystem; proper evaluation of cost for such customization must be done.
Another kind of customization can be for different market segments such as automotive segment which needs wide range of operating temperature and other environmental parameters.
In certain market segments like IoT, where the standards can vary by a large extent, SoC vendors prefer adding custom IP in-house rather than buying from outside.
Cost of Characterization – This is a big area where IP needs investment, specifically at advanced technology nodes where process can vary significantly between different foundries at the same node. It asks for the characterization of IP at every process variant. It depends how much pre-characterization can be done at the IP level; the SoC vendor might ask for special characterization at a specific node of choice for the SoC. A level of prudence can help here. An IP for GPU or mobile processing may need advanced nodes like 14nm FinFET and hence the characterization for process variants at those nodes will be needed. However, an IP for other applications which can stay at higher nodes may not need too many characterizations. But there may be other complications for specific applications. For example, an IP for automotive applications can stay at higher nodes (although moving down from 150nm and 90nm) such as 55nm, 40nm, or even 28nm for specific cases; however that will need characterization for a wide range of temperature and other PVT conditions.
Within an IP the characterization can be at the fundamental unit level such as bit cell and at macro level. The fundamental unit level characterization may not change frequently, but macro level characterization may change according to the design. So, that kind of characterization needs to be planned appropriately.
Cost of Acquisition – The cost of acquisition of IP is a very important aspect for SoC vendors. Large system companies have specific processes laid out for IP selection and procurement. They include items such as quality of the IP, ease of its integration, the IP vendor’s past record and ranking, vendor support throughout the SoC lifecycle, cost and RoI analysis as per single or multiple use of the IP, etc. It’s prudent to explicitly mention, specifically in single use, what kind of modifications and support elements such as error code revisions, defect fixes, configuration modifications, and so on are permitted. Also, the fees applicable for reusing and making variants of IP must be explicitly mentioned.
The evaluation of an IP and its integration into SoC is co-ordinated with the associated EDA vendors, development partners, and design service providers along with the IP provider. It’s a costly affair and hence it’s required that the list of selected and qualified vendors is kept short. The emergence of eSilicon as an IP service provider is a step in the right direction for IP evaluation before its acquisition.
An important aspect comes into picture when the IP needs some customization. In this event it’s important for both the IP provider and the SoC vendor to determine how the customized code will be maintained in future, whether the changes are generic enough to be merged into the main code branch. If not, then special support for that custom IP branch will be needed, asking for extra support resources borne by either the IP provider or the SoC vendor. So, here the question comes, how much is the support? Is it scalable and profitable for the IP provider to take it in her/his main stream? If not, then is it justified for the SoC vendor to acquire the commercial IP, customize it, and maintain it, or otherwise develop her/his own IP? If it is customization by the SoC vendor on top of the commercial IP, then the ownership rights must be clarified at the time of acquisition.
The cost of acquisition can also be factored in on a long-run production basis where the IP provider is paid on the basis of royalty fees. This can be finalized on the basis of specific terms such as actual sales or shipments. For an IP provider as well as the SoC vendor a typical challenge appears when the fab does not see enough RoI in creating a slot for a particular IP; this needs right level of negotiation before embarking on the journey.
I will park this part of the article at this stage as it has gone long. Stay tuned to read part – 2 of this article where I will mention about rest of the costs involved in using IP in modern SoC ecosystem. In the pursuit of increasing value of IP for SoCs to continue to have differentiated value, these costs must be understood well and accounted for in order to create a win-win situation between the IP provider and SoC vendor. This is required to maintain the IP industry at a healthy level and grow further from here.
Pawan Kumar Fangaria
Founder & President at www.fangarias.com
SPIE Monterey- ASML, INTC – High NA Readiness- Bigger Masks/Smaller Features