Chip design is, almost by definition, a matter of creating something new and different and unique. No two ASICs or systems-on-a-chip (SOC) are the same and there's little in common between the thousands of such products created annually. Likewise, there's little that the thousands of ASIC or SOC designers share in common. Everyone utilizes their own unique combination of favorite tools, programming languages, microprocessors, and emulators.

Yet, ironically, every engineer ends up using the same few microprocessors that every other engineer is using. Those microprocessors, designed in the main for RISC workstations of 15 years ago, have peculiar features and reduced capabilities which are now pressed into service in embedded applications for networking, media processing, consumer electronics, or industrial control. In fact, some of today's most popular 32-bit processors were used to power desktop Unix systems more than a decade ago-Unix workstations having very little in common with network routers, set-top boxes, digital cameras, video games, or cellular telephones, as any programmer will tell you. Yet the programmer's job is still to make the design at hand fit the fixed resources of a microprocessor that was designed with extremely different tradeoffs in mind.

In just the past few years, the law forcing compromise on off-the-shelf processors has been totally repealed with the advent of user-configurable processor cores. It's now possible to design a processor specifically for the system of the moment, with its own particular set of requirements.

If all aspects of the user-configurable processor-the programmers' model (register set, condition codes, and memory map), the instruction set, and the bus interfaces-are under the control of designers, then they can literally create the ideal processor for their task and for their task alone. This freedom to design a custom, perfectly suited microprocessor can be alternately frightening and thrilling. Frightening, because few engineers have ever designed a processor before. Thrilling, because the advantages can be huge: 10x improvements in performance are not uncommon; code size can be slashed with a "tuned" instruction set; and power consumption and die size can be cut to include only the needed features. Instead of writing code to fit their processors, engineers can now create processors to fit their code.

Another key advantage of a user-defined microprocessor lies in the fact that the user's customization is proprietary. OEMs own the instructions and other enhancements they have created; the intellectual property is theirs, and they share it with no one. This contrasts starkly with the usage of other licensed cores, IP, standalone processors, or DSPs, where every competitor is using the same processor or hardware architecture.

When OEMs own their own custom processor IP, several additional advantages accrue. Differentiating a product based on a broadly licensed core is tough. But product differentiation becomes much simpler when the product's architecture has been uniquely designed at the processor level including the register set, the instruction set, and bus interfaces. Rapid changes to that design are also possible when standards change or new protocols must be integrated. In addition, competitors cannot reverse-engineer a product based on a processor with custom features.

Owning the processor IP provides future proofing against the nebulous nature of the market and of IP vendors. Owning a custom-designed processor means OEMs can get their processors produced by the silicon provider of their choice based on price, location, turn-around time, reliability, power consumption, or process technology. They can also migrate their chips to a new process technology as soon as it becomes available, instead of waiting on the chip vendor's schedule. In addition, OEMs can control the production schedule, and therefore the product's lifetime, so there are no product end-of-life issues.

Finally, ownership of the processor IP means that OEMs are not dependent on the survival of the IP provider. In particular, OEMs who own the hardware source code are less dependent on technical support. Key pieces of silicon IP should be licensed from third parties to speed product development—but that shouldn't mean turning over control and ownership to another company.