IP vendors have long maintained that theirs is a product business model. In the face of rising complexity and shrinking geometries, however, selling IP may no longer be about baking up a plate of brownies, sealing it in shrink wrap, and putting it out there for sale.
IP is about creating a core of some sort, finding someone who needs the approximate features of that core, and then working closely with the customer to tune, tamper, and tweak it, until it fits exactly into a design. That requires services, and services are never shrink-wrapped.
The IP business model is not just about products today; it's a hybrid of product and services. But for every IP vendor who admits to the hybrid model, there's a baker's dozen out there who still swear they only do products, with just a dollop of service thrown in for good measure. Do these guys need to wake up and smell the baking powder?
I put that question to Ramana Jampala, President and CEO at CebaTech, in a recent phone call.
He said, "We provide tunable IP and, therefore, we're not afraid to use the word services. Services are an absolute essential in our business, especially for customers who are trying to push the envelope. CebaTech strongly believes that IP cores can add only so much value [to a project], because you also need customization of the core to [achieve] the required performance. It's through that combination of cores and services that we provide a way for our customers to stitch together the subsystems in their designs."
CebaTech started out life as an EDA vendor, selling a high-level synthesis tool to generate RTL from un-timed ANSI C. Jampala said, "The technology had application in everything from architectural exploration, to RTL generation and verification, and we had initial success in beta trials and universities. But selling our tool to RTL designers met with high levels of skepticism, so in order to prove what the tool could do, our team had to undertake the design effort itself. Producing IP as a proof of concept led to an evolution in our business model and a seamless continuum to today, where fully 80 percent of our revenue is driven by cores generated by our compiler.
"Our main expertise is in data networking and storage. With that expertise we've been able to use our ESL tool to deploy data management functions in hardware blocks in one-third the development time, without sacrificing performance. In the process, we have come to believe that basic core configurability is necessary, but not sufficient. Our customers need help sorting out the tradeoffs in design between footprint, compression, and throughput, and that requires services."
Joe Rash, CebaTech VP of Marketing & Business Development, was also on the call and added, "An ASIC design cycle can be 18 months to 2 years, but markets move at a dynamic pace. A shrink-wrap core with a pre-set number of configurations is not sufficient in an environment where customers must adapt quickly and make changes throughout the design cycle. Our cores and services together give customers the flexibility they need, from day one of a project all the way through to completion."
CebaTech clearly agreed with my thesis, the IP business model is more than a bake sale these days. For confirmation, I called IP veteran Hal Barbour, CEO at CAST, but got an adamant push-back instead.
For Barbour there is no ambiguity, IP is a product business: "There are a lot of companies who present themselves as IP companies, but they're really doing design services. There's a world of difference there.
"CAST is a true IP company, with most of our cores built around standards. We develop a C model as an extension of the design specification, and then code the RTL in either Verilog or VHDL. Once we've got that base code, however, we're only 25 percent of the way because there's still the vigorous process of verification, building testbenches and doing code coverage. Next, we put the design into silicon in some type of breadboard to make sure it's working, followed by extensive documentation. Finally, there's integration.
"Of course, an IP vendor also has to support all major industry backend processes. We have to have the tools and knowledge to make runs of the block in representative technologies from the various foundries. Also, if we've got a 90-nanometer run of a block, but somebody needs it for 65 nanometers, we have to know how to estimate what size it will be in the preferred technology. We also have to support the people working in FPGAs.
"It's just a helluva lot of work to develop and support a wide range of standards-based, bulletproof IP, and that's a very different business than design services. Anybody can put up a datasheet and call themselves an IP company, but they probably won't be providing the numbers that state this particular core consumes those resources and is compatible with these technologies. We would do a disservice to our customers if we didn't present them with these details, and the size and gate count for every core we sell."
Barbour continued, "Early on, when the IP industry was very young, almost all companies producing IP were actually doing design services. But today, if a company is not providing shrink-wrap IP, it's safe to say they're probably not an IP company. Yes, some IP does require enhancement and modification, and involves a technical engagement to nail down the differences between a standard product and what the customer needs, but the proper starting point is still a well-proven, validated foundation with a standards-based IP product."
Following my conversation with Hal Barbour, I spoke with Michel Tabusse, CEO of Satin IP Technologies. He carefully explained that his company does not sell IP, but sells software that helps designers and managers track and document IP quality metrics.
Tabusse said, "Our product is an addition to the huge and very complex design flow that our customers already have in place from tool vendors like Synopsys, Cadence, Mentor, and Atrenta. That huge flow creates megabytes of data after a block of IP has been integrated into the design.
"As an engineer, you have to scrutinize that data carefully. Our product retrieves valuable information from the design logs to show where you stand with respect to your quality objectives. The tool also helps IP vendors check how their core is doing and prepare for design reviews."
Tabusse said working with his customers has given him a perspective on the reality of using IP cores in SOC designs: "Methodologies for reuse may become standard throughout a corporation, but IP never is. Outside of very, very standard blocks of IP like USB or PCI controllers, I don't think it is feasible to view IP as a final product with no need for services. In real life, engineers modify and reconfigure the IP blocks they develop for SOC integration all of the time. They need help from their IP vendors to do that work."
Why does it matter if IP is a product business, a bake sale, or a hybrid of product and services? It matters because of setting expectations for customers.
If an IP vendor sells shrink-wrap cores that turns out to need significant services to implement, the customer goes away mad. Even worse, they overshoot their product development schedule. Alternatively, if an IP vendor overstates the need for services for a core that could easily have been deployed without it, again the customer goes away mad. Even worse, they overshoot their product development budget.
Either way, it's a lose-lose situation for the IP vendor and the customer, with only one solution. Total honesty from the get-go.
If you're an IP vendor and think your brownies are the best in the business, go ahead and bake away. Get out the plastic wrap, seal them up, and send them out. If, however, your brownies are better baked to order with or without chocolate chips, with or without walnuts, chewy or firm talk to your customers first. Let them help you decide which recipe and ingredients to use, and how they want it packaged up after the fact.
Either way, it's a win-win situation for the IP vendor and the customer. Total chocolatey goodness from the get-go.