Published on March 19th, 2012
Solving the SoC Design Enigma with IP
Increasingly semiconductor companies are experiencing the intricate design challenges of building next-generation SoCs.
Increasingly, semiconductor companies are experiencing the intricate design challenges of building next-generation SoCs. It's a fact that SoCs have quickly moved to advanced process nodes with many new chips being targeted for the 28nm process. At this node, there is a great deal of functionality that can and must be built into the SoC for the product to be competitive and have real market impact. The features that consumers demand in today's hottest mobile devices, for example--including smart phones and tablets--require a high degree of concurrent functions such as voice, video, graphics and web browsing for starters. This level of functionality requires that many processor cores, GPUs, DSPs, custom processors, DMA engines, and a host of peripherals be integrated into a single SoC. With the exception of the most vertically integrated companies, it is nearly impossible to own, manage and seamlessly integrate all this intellectual property (IP).
Frank Ferro is the director of marketing for Sonics and is responsible for the company’s complete portfolio of on-chip network IP products. Mr. Ferro has more than 25 years of experience in the semiconductor industry and has worked extensively in the communications and consumer electronics fields—with expertise in the areas of WLAN, Voice over IP, cellular phones, personal computers and SoC architectures.
Since companies can't build all their own IP, they must focus internal investments on technology that will give them a unique differentiation in the market and team with strong design partners to license the remaining pieces. It has often been quoted that for every dollar of IP that is licensed it costs about two to five dollars to use. So clearly a large part of the “make vs. buy” decision is finding well qualified IP vendors that will help to effectively minimize the cost and time of IP integration. Once a buy decision is made, there is a lot of work to do and questions to ask:
1)What is the maturity of the IP (i.e. quality/reliability)?
2) How will it perform in the target process node?
3) How can IP blocks from multiple vendors be easily integrated?
4) What is the business model?
Fortunately IP, design tools suppliers and foundries are getting a firm grasp on the challenges that customers face when trying to specify and assemble complex SoCs. Several companies are starting to work together to create a more seamless flow from IP selection, system design and back-end manufacturing. A recent example of this is the soft IP Quality Initiative being driven by TSMC and Atrenta. They are working with 10 IP suppliers as part of a soft IP alliance program to qualify the RTL deliverables of the suppliers. This initiative gives SoC developers a true peace of mind along with time-to-market advantages knowing that IP has been qualified and run through rigorous steps at the RTL level. This is a critical step, but is it enough for the overall SoC creation? Assuming that you did a solid job selecting well qualified IP, the next challenge is having the ability to integrate all of these disparate IP blocks and use them in a way that will deliver the maximum system performance and efficiency.
A critical IP block that is now at the center of the make vs. by decision is the on-chip network, or fabric, that connects all of these IP blocks together. SoC design companies are finally concluding that connecting 40, 80 or well over 100 cores together is a major challenge to design and verify in a timely manner. Difficulties stem from the fact that each IP core is likely to have a different interface protocol (AMBA, OCP, proprietary, etc.), different bit widths, different frequency and different power requirements.
In addition to connectivity, the performance of the overall system is greatly impacted by the design of the network since all of these heterogeneous processor cores will be competing for memory bandwidth. The network will have to deal with the overall quality of service, error management, security and even system coherency in some of the more advanced designs. Selecting the right network with all of these features is an essential step to ensure the system meets the stated performance goals and can execute on schedule. Since the on-chip network is the ‘nervous system’ of the SoC, with the correct design tools, this is not only a critical IP block, but a platform methodology that will allow SoC designs to scale as requirements change--ensuring the fastest time-to-market with incremental and new designs.
So are advanced SoCs difficult to design? Yes, as you can see from some of the challenges outlined. However, real strides are being made by IP and tools manufacturers to assure the quality of the IP. In addition to quality, adopting an on-chip network platform methodology is becoming a key part of executing a successful SoC strategy -- allowing any third party IP from any source to be easily connected to the network. This methodology will help reduce that two to five dollar cost of IP integration and help companies remain competitive in the fast growing markets that require complex SoCs.