Today’s guest blog is from Mark Gogolewski, who is one of the founders of Denali, famous for
parties memory controller and other IP. Denali was founded 14 years ago and was financed on the sweat-equity of its founders who initially paid themselves nothing and is today still privately held (which is another story I’m sure) and about 150 people. They don’t release their financials but are widely believed to be very profitable.
Stop the insanity
Massive amounts of virtual print have been dedicated to the negative implications of the “all-you-can-eat” EDA bundle. I share this view, and hope (actually, expect) that this downturn will trigger the kind of creative destruction required to force at least the struggling major players in our industry to change.
In the meantime, the insanity continues (Einstein’s definition, not the psychiatrist’s). The EDA bundling virus has extended into IP. We are being hypnotized by promotions of design and verification IP bundles in the name of “complete solutions” and news of major players buying up smaller (usually unsuccessful) verification IP so yet larger bundles can be created.
An astute observer realizes that the all-you-can-eat IP bundle lacks even the fundamental promise of the EDA bundle: the value of integrated flow vs a collection of best-in-class point tools. Self-serving as it is, we at Denali believe IP is different. Of course, integration is important, but standard on-chip interfaces make this a 2nd order criterion. Some problems are complicated enough and important enough to require customers to pick the best in order to properly mitigate risk and maximize performance.
That being said, and with apologies to Mr. Letterman, I offer the Top 10 reasons the future of IP business is not all-you-can-eat.
- 10 No company can have an unlimited set of core competencies. Each major type of IP requires a distinctive core competency. That is, there is special sauce for making CPUs vs embedded memory vs DRAM controllers vs analog components vs PHYs. Realistically, companies have only a few true core competencies that drive product differentiation, success and ultimately profits. Is it realistic to believe one vendor can have them all? Should we expect the same behavior to lead to a different result?
- 9 IP is a treadmill; EDA is plant-and-harvest. Lucio Lanza has made the treadmill point many times. Protocol-based IP is constantly changing – and doing so quickly. PCIe goes Gen 1, 2, 3. DRAM moves to DDR1, 2, 3, while also splitting to LP DDR and LP DDR2. Etc., etc. For EDA, you make a major investment up front and then reap the rewards for a decade. There is a fundamental impedance mismatch between execution of IP and EDA tools. Even the sales momentum is different. If someone chooses your place and route they are likely to be using it in a couple of years. If they buy your USB1 core then they may or may not buy your USB2 as there is less connection. You not only have to keep developing, you have to keep selling.
- 8 Name a major IP product selling today that was developed inside a large EDA company. Yeah, I can’t either.
- 7 Relative revenue leads to relative focus. Many IP markets are $5-20M in annual revenue. What focus can/should you get from a company where the product represents <=2% of revenue?
- 6 Large EDA suppliers represent an Axis of Evil. OK, just kidding.
- 5 EDA OEMs never work . History shows IP OEMs are no more successful.
- 4 One IP product alone requires 2 core competencies, maybe 3. In order to make a piece of IP a success, you need at least two things. The first and the most obvious is you need to architect it right. With a strong micro-architecture, getting the design coded is more straightforward. Next you need to verify it. Verifying IP requires VIP that fully covers the protocol, can be configured for all needed BFMs, checks for every error, and allows for all relevant errors to be injected. For example for PCIe, the IP itself has to respond to errors across the interface in the proper way. So there can be as much or more "negative-stimulus" testing as there is proper functional testing. Now the third leg of this is that once you decide to make IP for the world, you need to optimize how to build, test and deliver customer specific IP again and again and again. Inside Denali, we speak of this as the need for differentiated infrastructure to deliver both product quality as well as operational efficiency. Without it, your product fails or your P&L fails.
- 3 Free code does not belong on your chip. For a buyer, a major allure of bundling is getting things thrown in for free. Outside of really simplistic problems (UART, bridges), how many pieces of free IP do you want on your chip? How much free VIP should you use to verify functionality and compliance?
- 2 Must pay the EDA sales channel piper. Successful public IP vendors have cost of selling on the order of half that of EDA. Can a customer afford to have so much $ go into vendor sales vs renewing R&D, Verification and Support? Can the vendor?
- 1 Synopsys currently is the exception to everything I said…which proves the rule.