Peggy Aycinena is a freelance journalist and Editor of EDA Confidential at www.aycinena.com. She can be reached at peggy at aycinena dot com.
Hal Barbour: Master of the mega-trend
August 30th, 2012 by Peggy Aycinena
Hal Barbour is President of CAST, an IP company based on the East Coast. Hal has a tremendous ability to explain the many facets of the industry, and it was a great pleasure to sit down and talk with him this week. When we spoke by phone on August 29th, he had just wrapped up an earlier call with a customer.
Q: How do you make yourself known to customers?
Hal Barbour: We have always put a lot of information in the hands of our customers, but the delivery mechanism today is quite a bit different. We’ve learned to leverage most of the contemporary tools – blogs, online meetings, webinars, shows and press releases. Press releases are just as important as ever, but where we used to send them to a central distribution center and a group of editors, now there are about 15 or 20 various people and outlets who disseminate the information to a much larger population.
Q: And how do working engineers hear about the products?
Hal Barbour: That’s the really interesting thing. Engineers today can easily see press releases directly, plus they have at their disposal a powerful set of search tools to help them get the information they need, so whatever information you’re putting out there, it better be right and it better be credible. If it’s not, engineers have got plenty of other sources to turn to.
And if you’re going to be out there, you better be able to respond to inquiries quickly and rapidly. Ultimately, however, it’s your name and your reputation that sells products. I can’t tell you the number of people who contact us based on our name and reputation.
Q: Isn’t that called ‘word of mouth’?
Hal Barbour: That’s exactly what it is, only it’s even faster today. Spreading the word used to be limited by who you knew, but today with social media and blogs, word of mouth moves at lighting speed and is more important than ever. Even today, though, nothing substitutes for face-to-face contact with the customer.
Q: You mean the old fashioned face-to-face sales call?
Hal Barbour: Absolutely. Email and Skype can never substitute for the personal sales contact. We’re always trying to fine-tune our sales process. We have marvelous people with complementary skills, sales people and technical people who understand the message of the customer.
We may have 6 different CAST people in contact with a customer and want to keep all lines of communication open – which is challenging – so we’re always working to improve the process.
Q: If somebody calls a trusted colleague, asks for the name of an IP provider and is told to call you, but you don’t have the product they’re looking for, what happens?
Hal Barbour: If we don’t have it, we steer them in the right direction to someone who does have it. After 20 years in the business, we’re not very cautious about giving out recommendations for people we have dealt with before. We try to provide the kind of service and help that we ourselves would want as customers.
In fact, at one point we were part of Warren Savage’s Constellations program, but as our businesses evolved we decided not to get into a competitive situation with Warren – both IPextreme and CAST deal in 32-bit processors. Nonetheless, I salute Warren for establishing the program, which is an excellent effort to collaborate with complementary business partners.
Q: Do you use Sales Force-type products to keep track of your customers?
Hal Barbour: Sales Force is a tool that would probably be our tool of choice today. But more than a dozen years ago, knowing it would be important to have a customer database, we decided to develop a database system of our own built on Lotus Notes.
It’s the original, customizable, collaborative, working document management system that was initially used by groups in large companies to manage things such as project reports, travel reports, expense reports, performance reviews, and so on. But it could also be used to manage customer information, so we created our own Notes-based CAST system for tracking all our prospect and customer info.
Today, we’ve got upwards of 40,000 records in our database, with over 100,000 contacts. Our database of customers and contacts is important, particularly as we have about 1300 cores licensed out to going-on 800 companies.
Q: Speaking of licensing, what’s the business model at CAST?
Hal Barbour: There are a variety of different license models, and those differences are identified at the outset of our engagement with the company.
Mostly, we charge an up-front fee for use, a project-based license. When a customer licenses a core, it’s for use in a specific chip. They are then allowed to produce as many chips as they want that include that core. But if a customer is going to do multiple chips with that core, then we develop a different license for each special case. Sometimes it’s multi-use, sometimes subscription, and sometimes it’s a yearly fee.
Q: ARM generally goes with a royalty model. Does that impact your business model thinking?
Hal Barbour: Yes, we do have to track what ARM is doing, but we’re not constantly competing with ARM even though we certainly have a superb range of 32-bit processors. Our high-performance 32-bit core is more targeted at deeply embedded applications. Our customer’s customer usually won’t be programming the devices themselves, so the ARM brand name is not a limiting factor.
Within ARM’s infrastructure, a customer is seldom going to venture outside of ARM’s umbrella, but there are a whole host of applications where the customer doesn’t want that. Instead, they prefer a business model more like an 8-bit machine, like the 8051, a place where we are the world leader.
Q: There’s still a big market for 8-bit processors?
Hal Barbour: That’s the beauty of the 8051. It’s well known and comes with a host of established development software tools. At this point, there are probably 200-to-300 chips out there that have been fabbed using our 8051.
Q: If a chip containing one of your cores is super successful, do you lose sleep with regret over not having engaged with that customer using a royalty model?
Hal Barbour: Yes, sometimes we wish there had been a royalty. For example, one of the world’s leading smartphones has one or more of our cores on board. This company has been highly successful using chips produced in the multi-millions of units, all incorporating CAST cores. Sure, it would have been nice to have had royalties there, but the 8051 is in the public domain. It’s not a proprietary design like the ARM processor, so it was not realistic to expect royalty licensing.
Q: Do you ever have concerns about the impact of problems in a design that might include your core?
Hal Barbour: We’re proud to say that in over 20 years of doing business, we have never had a lawsuit. Certainly CAST and every other IP company is concerned about lawsuits, about limiting liability to the value of the core itself, but sometimes the customer wants higher levels of indemnification. We’re not willing, however, to go higher with that than the value of the core itself.
In this area, the work of the GSA IP Working Group, in which we actively participate, has been very useful. We have polled a number of users and the IP companies that supply cores, and we can see from that work that the standard liability is 1 or 2x of the core value, and no higher. Perhaps some IP vendors provide higher indemnification, but our customers have usually accept our position – after they have seen the GSA report – that the IP licensing terms we offer are industry standard.
Q: Regarding confidentially: If I’m a potential CAST customer, how do I tell you what I need, without really telling you what I need?
Hal Barbour: There’s usually an NDA in place, and it’s our policy to use the one provided by the customer. We found that if we provide the NDA, it can take weeks for the customer to run it through their legal review, so we use the customer’s. They send it to us, we review it and sign it, and the conversation begins. We need to understand what the requirements are about, what the customer is trying to do with the project and the sooner we get to that point, the better it is for everyone involved..
Q: At what point do the design engineers need to talk to the IP vendor engineers?
Hal Barbour: We are not like a lot of other companies with people with soft skills who interact with potential customers. All of our sales people are really good technically. The customer can submit a detailed list of questions and generally receive answers within a 24-hour turnaround time.
CAST has closely-knit relationships with developers in Greece, Poland, and Germany, plus our own team in Czechoslovakia, as well as employees and partners in North America and Brazil.
The time difference between these places works to our advantage in getting back to the customer quickly with answers. If we pass a set of questions on to our developers at the end of the day, generally when we get up in the morning, the answers are on our desk and we can send them on to the customer.
Q: Do you ever sleep?
Hal Barbour: [laughing] No never. Our people really enjoy what they do and it’s not unusual to find CAST people on line from the early morning hours to late evening hours.
Bear in mind, we have taken no outside investment or debt at CAST. The owners are all employees. Our first products were simulation libraries, which then morphed into IP cores – before the industry knew that term – with the difference between a simulation model and synthesizable core being the construct that allows it to be synthesized.
As IP began to grow, we knew we needed to add more people, so we have chosen partners in other parts of the world where there are excellent educational institutions and trained EE’s but that don’t have particularly good access to global markets. It’s the global nature of our business that really compliments their IP development skills.
About 15 years ago, we found a group in Poland, Evatronix S.A., who had the background and skills to do what we needed, and the cost of an engineer in Poland at that time was about 15% of the cost in North America. That gap has since closed somewhat, but it’s still a lot less expensive for extremely high quality work.
In Greece, we have Alma Technologies. In Germany, Fraunhofer provides our automotive controllers. We have our own development center in the Czech Republic, and about 2 years ago added our latest development partner in Slovenia, Beyond Semiconductor. Beyond are experts in front-to-back design and have their own proprietary 32-bit processor, the BA22, which is embedded in quite a few tier-1 company designs. We also have long term distribution partners in Israel and Asia. Plus, we also have our own offices in China.
CAST focuses on shrink-wrapped IP, so we often partner with these different developers, take their silicon-proven IP, and make it customer-friendly and easier to use. We’ve tried to be wise in selecting our partners, leveraging their intelligence and backgrounds to achieve a lower-cost business model. We’re pretty frugal here in North America too, keeping our overhead to an absolute minimum.
Of course, we need very experienced people to pull it off and we’re one of the few companies in the IP or EDA industry that has succeeded here. It’s not for the beginner or novice – but like JetBlue, where many people work from home unless they’re servicing or flying the plane, we’re champions of the home-office model.
Q: Looking out across the IP landscape, who are your competitors?
Hal Barbour: Our cornerstone is the 8051 and in that area, the competitors are in France and Poland. At the higher end, clearly there’s ARM and MIPS, Synopsys and Tensilica. But we’re going at it in a slightly different way, through deeply embedded applications where the business model is more like the 8051 licensing. There may be a higher price for the high-end processor, but the business is similar in many ways.
In general, our customers liked our 8-bit machine in the past and now need more performance. If you can create very compact code that needs less memory, and include better characterization, you can compete very effectively. That’s why the BA22 processor family we offer is going into some incredibly high-volume applications. There’s plenty of room in the business out there.
Q: Once purchased, how is the CAST product delivered?
Hal Barbour: The delivery mechanism is almost always over the Internet. The customer needs to provide two things – a signed license agreement and a purchase order. We then typically deliver in a few days via a protected download.
Usually an email goes out to the customer with a link to an FTP server. As soon as the customer goes in and gets the code, it’s deleted from the server. Then it’s ready to be unwrapped. There’s a TAR file, the customer contacts our support organization, and they’re up and running. It’s a very straightforward process.
Q: Could you have guessed when you started 20 years ago, that it would come to this?
Hal Barbour: No, initially it almost always involved physical shipments. But with the Internet, that has all changed. In Asia, sometimes there is still physical delivery of a CD, but that’s only in Asia – perhaps China, or Taiwan. But in North America, I can’t remember the last time we’ve made a physical shipment, with the exception of hardware debugging pods for the processor cores.
Q: On a more serious note, how can you trust that your IP products are only being used as proscribed in the license agreement, that they’re not being used as collateral in a poker game or given away to friends?
Hal Barbour: We believe that if you have to go to exotic means to make sure that IP is so secure it will never be used illegally, you create such a complicated morass of steps and procedures that both from the supplier’s point of view, and that of the customer, everybody suffers.
In my previous life, at an EDA company, some of the tools we were licensing were from a group in Europe, and that group was positively convinced that absolutely everybody was trying to rip off their place-and-route tools. To authorize the use of the product, the customer had to flawlessly enter between 80 and 100 characters without a single error – a string of characters that included numbers, and upper and lower-case letters. One error, and they had to start all over again from scratch. It was a nightmare and we vowed that we would never again get into a situation that required such complicated procedures. At CAST, we rely on the license agreement as the primary vehicle to prevent misuse.
When you’re designing a chip, particularly if it’s in a contemporary technology, the downstream costs of generating the masks and manufacturing the chip are so high compared to the cost of the IP – who is going to risk those kinds of costs with a questionable piece of IP? If they did and the chip somehow failed, the penalty for getting caught would just be way too high. And frankly, most technology businesses are pretty honest so the only threat really is inadvertent use, rather than conscious misuse.
In addition, no matter how well tested and documented the IP, it’s still a complicated process to integrate it into a design. Often engineers using the IP need a small change. They need access in those cases to the engineer behind the IP, someone very knowledgeable.
To rip off a piece of IP and use it without technical support can generate penalties that are beyond acceptable. The penalties for getting it wrong are enormous!
Q: So the entire infrastructure around IP plays to the honesty of the users?
Hal Barbour: Absolutely.
Q: And your faith in humanity has not hurt the business?
Hal Barbour: No, our faith in humanity has not hurt us at all. Those without faith in the user community are totally detached from the practical issues involved in designing around a piece of IP, the cost of getting it wrong versus the rewards of getting it right.
Q: How much of the cost of a chip is consumed by the IP?
Hal Barbour: It depends on exactly what kind of IP you’re talking about. If you’re using proprietary IP from Cadence/Denali, Synopsys or ARM – it’s just a guess, but at the very high end, you’re talking about no more than a few cents on the dollar. It’s really not a lot.
The real cost in the chip is in laying it out, in the mask set, in the production – particularly as we go to smaller and smaller geometries. When you get down to 28 nanometers, the overall projects costs are 100x what they were at 65-or-90 nanometers. So IP may cost you, but the real costs are elsewhere.
Again, a large percentage of what CAST does is standards-based digital IP. Many of today’s chips utilize high-speed serial buses, and may include standards such as PCI Express, USB, Serial ATA, and so on. In these designs, the real cost is not in the digital part, but in the mixed-signal or analog bits.
A high-speed serial interface, done for a particular library, in a particular geometry that may also require hand crafting, is going to be expensive. A digital controller core may cost you $50,000 to $100,000, but the transmission/receiver side will cost you anywhere from $300,000 to $400,000 for the design alone.
Q: If a customer told you the team was suffering from NIH and not happy about buying 3rd-party IP, would you just laugh and tell the customer to get over it: It’s the 21st century.
Hal Barbour: The natural law of economics dictate how much NIH can be tolerated, while it’s the market that drives the need for IP.
Certainly there are still legitimate cases where people need to design it in-house – particularly in the area where the company’s secret sauce is centered – but it makes no sense whatsoever at this point to invest time and resources in developing things that are the same as what others are doing. USBs, serial ATAs, JPEGs, H.264, and so on – if you’ve got your engineers spending man-years developing and supporting this stuff, it’s going to cost you a lot more than using commercially available IP.
If you need a processor today, few can afford or justify the costs of developing their own IP cores, tools, compilers, or debuggers And it’s not just the development, but the maintenance costs as well. It’s just a helluva lot cheaper to go outside, and concentrate your resources instead on where you can truly differentiate your product.
Q: You really do a great job explaining all of this. You should be teaching a course called IP 101.
Hal Barbour: It’s true, I’m interested in the mega-trends. And what I’m seeing now is the next mega-trend emerging: The IP industry is migrating quickly to sub-systems, rather than simply IP blocks.
Having the understanding and experience to meet those demands, with the know-how to assimilate groups of blocks – in-house IP plus groups of IP from others – that’s the direction where the industry is going. And, that integration needs to include the necessary firmware, possibly a conglomeration of propriety IP and standards-based IP, as well as the tools to integrate it into the larger design.
Few companies can do it all from architecture, to detailed digital design, to physical design, to tape out, and then there’s the software and firmware. Clearly, the right partnerships are important to complement your core competency – no pun intended.
As far as explaining things: I like to look holistically at the big picture and how it’s being formed. Descriptions of the IP industry, for me, always have to start with the big picture. That’s what energizes me.