Balancing Quality, Cost And Locale

By Ann Steffora Mutschler
As more features are packed into a single SoC there are simply more time-critical decisions to make. Instead of holding up one chip of a six-chip chipset, a delay or error on one chip can stop the whole parade.

That explains why one of the most vibrant parts of the business at big EDA companies these days is standard IP, and why most of the other commercial IP makers are issuing bullish forward-looking statements. IP is only a piece of it, too. With more and more companies looking to maximize their resources in building increasingly complex chips, whatever can be outsourced is a win—sometimes even the design itself.

“No matter what you are outsourcing, whether it is payroll, IT or semiconductor operations, the argument against it is always control,” said eSilicon chairman and CEO Jack Harding. “That’s a fallacy. There’s virtually no company in the world that does its own payroll, yet they control it fine. There is virtually no major corporation in the United States that manages its own IT yet they control it fine. In the case of our customers that do semiconductor operations outsourcing, it’s the existing operations team that typically says they really don’t want it to happen – it’s more self-preservation.”

SoC or PCB?
In terms of SoCs, that world is coming under a lot of pressure between issues of price point, time to market, quality, and schedule predictability, said Charlie Janac, chairman and CEO of Arteris. The impact is that increasingly, SoCs are going to start to look like PC boards because they will be assembled out of pre-made parts. “There are some parts that are not going to be like that and those parts are where the company makes the IP internally, which is reflecting its competitive advantage.”

The rest can be bought. “That makes a lot of sense because to build an internal IP is becoming very, very expensive. The verification is the main cost, and if those costs can be shared across the entire industry, then that allows the SoC maker to deliver that SoC at a more attractive price point. It’s not about technical capability, particularly with the large SoC makers, but it is about running a profitable business.”

Whether or not the IP is a differentiator for the semiconductor company can also determine where IP will be acquired from. When it comes specifically to IP models, Ran Avinun, marketing group director for the system design and verification segment at Cadence, believes that customers want them from the IP provider, not the EDA tool provider.

“They are looking at anything that is not their core competency as something they are willing to get from the EDA companies. This could include some of the memory models, or some of the I/O subsystem models,” he said. “Semiconductor companies don’t want to, and simply cannot, produce all of the IP models by themselves. They don’t have the bandwidth or resources to do it. As the real estate in each ASIC becomes enormous, they can’t do it by themselves. They have to rely on their ecosystem, and they prefer to focus on their core competency, which is IP that is unique for their business, or their IP stack – they way they connect to peripherals or software layers,” Avinun said. “We’ve seen that there is a shift in many companies. In the past, they used to focus on their IP creation, the process they are using, the fab. These days, a lot of those companies are losing those fabs or selling their fabs and they don’t want to develop all of the IP, so the name of the game becomes integration and how do you integrate all of these components with our combination of what you developed plus with what other companies are developing – integrated in an optimized way.”

You get what you pay for
In any situation, with any product, cost is a factor – which goes hand-in-hand with value.

“To go low cost and low skill doesn’t buy you a damn thing,” Harding said. “I would argue that it’s not the lowest cost you are seeking. It’s the highest value you’re seeking, which is output divided by cost.”

And interestingly, based on perception audits, eSilicon found that people will pay more for the company’s U.S.-based team – partly because they are very senior, but partly because they are based here. “We’ve seen over and over again with one competitive threat after another, none of the Asian-based competitors we have, they’ve all tried to land here and go upscale into our market and they can’t because their businesses are based around extremely low cost. The converse is true for us—we know how to make money with a relatively high cost basis by having higher quality engineers who get more yield per wafer and lower test times and smaller die sizes and the like – we are able now to go into Asia, transporting a methodology that works in a complex setting to a very low cost basis,” Harding noted.

Where IP, IP models or design services come from geographically is of concern to some, to many others it is not. However, a determining factor across the board in choosing an outsourcing partner is quality.

“I’ve seen a tremendous concern about quality and people want quality IP, no matter which part of the world it comes from,” Janac said.

Of course, quality IP means different things to different people. “From the SoC architect’s point of view, it means there are no architectural issues with a given IP. From the SoC assembly people point of view, it means that IP goes through the SoC integration and physical design process smoothly. From the director of the project point of view, it means there is no impact on schedule – to them bad quality means being late and having problems in the field. Basically, the more an IP has been verified in-real world designs, not just verified with tools, the more valuable it becomes and the more confidence people have in its quality,” he concluded.