A cold and changing world for silicon IP
At this point in the year it is customary to sit back, ponder the year’s events, and try to divine some trends: some patterns in the flocks of news releases, meeting notes, and conversations that now gather, take wing in the chill air, and turn South—out of the summer country of short-term memory and toward the cold land of yesterday. To be fair it is not from any improved perspective that we work this augury, but merely from the cold fact that in this bridging week there is so little news.
It need not be said, but to 2009: good riddance. This year has seen many fine engineers take up consulting. It has watched promising companies scratch for revenue, find none, and quietly begin shopping their IP around as the precious capital drips away.
But these hard economic facts may have obscured an entirely separate pattern: a technological shift in the IP industry. The visible surface of this shift has been consolidation. Beneath that level moves an even more pervasive change: increasing integration is pushing the need for system-level expertise further up the supply chain. SoC vendors have become systems vendors. And of necessity, IP vendors are having to assume responsibility for complete functional subsystems.
The pattern has been evident for some time. Application-independent IP blocks such as DRAM interfaces are now sold as complete functioning subsystems, not as sets of parts from an IP catalog. ARC and Tensilica, the two major configurable processor core vendors, have long since shifted from trying to sell CPU core kits to providing complete hardware/software/verification packages for specific functions, such as handset audio processing. The configurable processor core is still in there, but it’s almost incidental to the product. In fact the entire applications plane of an Android handset appears on its way to becoming a single piece of IP.
What does this mean for IP vendors? Fell words. Unless you are offering a full subsystem or an analog/mixed-signal function not available elsewhere, your customers will be other IP vendors, not SoC designers. And your margins will be nil. If you do offer a full subsystem, the cost of admission will be a fully-functional reference design in the target process—forget FPGA implementations—with all the software, including some differentiating functionality, a full verification suite, all necessary industry certifications, and the design support necessary to ensure your customer will succeed in tweaking and integrating your subsystem.
If this sounds beyond your means, there is a simple conclusion: you are too small to stay in the silicon IP game by yourself. Your business is equivalent to offering quad NAND gates to an industry that is designing with microprocessors. Not impossible—there are still vendors selling discrete logic today—but not exactly a business plan you would want to present to investors.
So welcome to 2010. If we are lucky, the current supply-chain mini-bubble will relax gracefully instead of bursting, leaving behind slow, excruciatingly margin-conscious growth punctuated by a few stand-out products areas. As the storm clears, we will look to the skies and see just how much of the IP business we knew has migrated to a new model, and how much has flown South forever.
DannyBoy commented:
H@UK commented:
Rich Wawrzyniak at Semico Research commented:
garydpdx commented:
Phil Casini - Advance Tech Marketing commented:
Jüri Põldre commented:
IP commented:
Dieter ERNST commented:
shawnkempf commented:
