What Would Joe Do?
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.
IP Update: Stamme @ Kilopass
March 10th, 2012 by Peggy Aycinena
Even this deep into the era of IP and design reuse, it’s been my impression that things are not quite as far along as many in the industry would like you to believe. With that attitude in mind, I spoke with Bernd Stamme, Director of Marketing & Applications at Kilopass Technology, who convinced me otherwise, although with several caveats.
Q: What kind of IP are people buying these days?
Bernd Stamme: They’re buying everything that has to do with performance – computer performance, networking performance, and elements that control that performance.
The one way you can distinguish yourself from the competition is to go faster. You can use IP for CPUs or GPUs, or off-the-shelf interfaces. At some point, you’ll run out of what you can get, however, and you’ll need to build auxiliary units around it to improve performance for certain operations. Then you’ll be building co-processors, specific circuits to accelerate [the system] and get better performance than your competition.
Q: How do you know what IP to use and when?
Stamme: You need to have a product plan that’s defined around your performance parameters, but prior to that you have to have a market plan. You need to know where you are now, where you are going to be, and define the market requirements of where you hope to go.
You need a budget for performance, how much power you want to burn, etc. It’s all part of a market requirements document [detailing] what you need schedule-wise and product-wise. From a commercial point of view, it’s what you need to have any chance of success.
Once you’ve got your requirements document, you [hand it to] Engineering and they come back with: This is what we can do in software and this is what we can do in hardware. We need a processor CPU for this and IP licenses for that, so we can implement a specific circuit to meets your requirements more efficiently.
Q: Is there optimization software to help maximize the partitioning between hardware and software, and help decide what to build in-house versus what to buy from third-party IP vendors?
Stamme: Actually, it’s like a restaurant [and not so easily optimized]. The customer comes in, places an order, and then expects a certain response time. The restaurant has to maintain the quality of the food, which is a complex process, allocating the right amount of resources in a given amount of time, and keeping the customers happy.
Q: How does Kilopass play into this equation?
Stamme: We’re part of the equation for the embedded non-volatile memory. If you need some data that’s residing on the chip, even if the power goes away, then our IP comes into play.
Q: What competitive advantage are you offering with your IP, and why shouldn’t I do it in-house?”
Stamme: Because Kilopass IP don’t require any additional process steps in the SoC. If you’re designing standard analog or digital chips, most non-volatile memory is not compatible. They require others things to make them work – additional, expensive process layers, or additional handling steps that add to the cost of chip manufacturing. Our products do not require any of these additional process steps.
Q: And it’s not possible for your customers to do that in-house?”
Stamme: It requires a lot of expertise that most customers don’t have. [It’s true] that as recently as 15 years ago, people would design their own RAM, ROM, and so forth, but today a very high percentage of all designs include some sort of memory IP from IP vendors.
Q: Nobody can do this in-house anymore?”
Stamme: Well, the foundries can do it. They can compile silicon. Or, the people who do standard processors like Intel or AMD, who need to get maximum performance out of their products. They bin their products in manufacturing, sorting them on the test floor for different speeds and price points. However, this is not an approach you can use with ASICs, because they need to follow the standard range.
Q: Therefore, IP reuse is in a good place these days?
Stamme: Well, of course there’s always going to be the concern: If I bring in IP, how many other players will be doing the same thing I want to do? Particularly if I pick an IP that a lot of other people are using.
Then there’s the issue of risk mitigation: If I do it myself, I can trust it, but if I take it from the outside, there’s always some risk associated with it.
For example, if you want a microprocessor core for a mobile app, most people choose ARM. So if I pick that one, for instance, I’ve got a warm and fuzzy. I know it’s a safe choice, because a lot of people are using it, banging on it, and trying to get the maximum performance out of it – all of which leads to improvements in the IP, and actually reduces my risk in using it in my design.
Q: Is there some level of usage, some critical mass of usage, that indicates this IP, or that, is appropriate and reliable?
Stamme: Yes, that’s often true. However it’s also the case that if there’s no competition, IP providers can get slow and sloppy. They get lazy and their products grow less and less competitive. Then the customers hate it, because the performance is no longer good. That’s when some other vendor jumps and up and says: You should use this other IP!
Q: How do you fight the 800-pound gorillas in the IP vendor world, if you want to sell IP?
Stamme: If your competitor’s products aren’t as expensive as yours, it’s tough to compete. Then you need to find an engineering manager in your potential customer’s organization who has run into problems with the Gorilla core – someone who’s discovered he can’t continue with the [received wisdom]. As an IP vendor, you have to see this as an opportunity to get your potential customer to make a change.
Of course, there are always hang-ups with IP – nobody is ever fully happy with what they get. If you look at any customer’s R&D project, you’ll probably see that there’s always a certain aspect of the core they’re using that they don’t like. If you can present them with a product that overcomes those limitations, you’ll be successful.
It’s never easy, however. In fact, it’s getting more complex with each technology – the on-chip contents are getting more complicated, by orders of magnitude, with each chip.
It’s like Legos. Originally, there were only 2 shapes and 4 colors. Now the options are mind-boggling, and Lego is the largest company in Denmark. Chips are the same. A long time ago, you had gates, but now you’ve got IP, often in terms of sub-systems that represent very complex chips within themselves.
It’s not getting easier as the complexity is getting bigger, so the choices are key. The price for making a wrong choice and wrong functionality goes up astronomically. If you screw up a 40- or 20-nanometer chip, that’s several tens of millions of dollars [down the drain].
Q: Why would anybody want to go into this business?”
Stamme: Actually, [perhaps because of that] we’re seeing a shift to Asia – more risk taking there, more short cuts to the final design, and lower price points where the cost of failure is far less.
For a whole host of analog chips, however, the real cost is not so much in the IP as in the expertise of the analog designer, so at this point, an analog company like Maxim is thriving. It’s clear there is still a lot of opportunity in this business!