Design in the era of ultra-deep submicron (UDSM) silicon is a constantly moving target. To hit the bullseye, you've got to fill many more gates per given amount of area, verify that much more functionality, and know that almost anything can go wrong when you're deep into the umpteenth simulation run.
An expanding palette of tools, methodologies, and languages can be combined in innovative ways to find the right solution. Among other things, the design challenges will call for improved use, and reuse, of IP. It'll also mean adopting some design methodologies that have been slow to see broad adoption, namely algorithmic design and coprocessor synthesis. Such technologies won't fix everything, but they're part of a broader solution.
Similarly, the verification puzzle must be addressed through a mix of elements that includes tools, languages, and smart planning to achieve adequate coverage.
THE DESIGN SIDE
Today's end-user markets are driven by increasingly integrated portable electronics, more often than not involving some kind of wireless connectivity. For designers, evolving industry standards in the multimedia and wireless domains means more digital signal and control processing.
The flexibility to integrate and adjust to these evolving standards has become critical. Plus, the increasing complexity of emerging algorithms drives the desire to implement more of the signal processing algorithms in software instead of hardware.
It's too costly, and risky, to go completely custom with the implementation of all of this portable functionality. So, designers need ways to implement reconfigurability.
Fortunately, several alternatives will begin to take off in the near term. For one thing, more designers are turning to FPGA and/or structured-ASIC technology. Second, platforms consisting of multiple custom processors enable complex algorithms to be implemented in software instead of hardware.
The customized instruction set of custom processors increases-software performance while keeping power consumption under control. The ability to spread the processing load across multiple processors also makes it possible to fully exploit the parallelism inherently present in DSP algorithms.
In addition, reconfigurable computing combines the reconfigurable logic and custom processor into a reconfigurable processor. Such processors consist of a processor core that's tightly linked to an array of reconfigurable computing elements that at run-time are (re-)configured to accelerate critical software.
All of the above is only possible through use of ESL tools and methodologies. They ultimately provide the required programming, algorithm, and architecture design tools.
On the algorithmic side, the industry will start to see IP offered in conjunction with algorithmic synthesis environments that effectively explore the design-space alternatives. No longer can you just trade off area and frequency as in RTL synthesis.
Rather, system-level parameters will be employed to trade off power, sample rates, and system throughput in addition to area and frequency. Combining algorithmic synthesis and IP hardware will reduce development times and integration issues, as well as increase the use of third-party IP.
We'll also see more use of application-specific compute engines to address worsening SoC performance and power-consumption problems. The most successful design technologies will offer an " incrementalist" approach that preserves existing design investment and infrastructure.
Standard processors alone don't have enough hardware parallelism to execute compute-intensive SoC application software in a power-efficient manner. Thus, design teams are proactively evaluating technologies for developing application-specific computing engines to boost parallelism.
Look for emerging new concepts and applications of IP reuse, for both design and verification IP, to support SoC methodologies in 2006. These next-generation reuse methodologies embrace the concept that for reuse to be effective, it must bring together all aspects of IP. These new methodologies treat software, hardware, verification, and documentation with equal respect.
Traditional approaches generally use text specification as the golden reference, with all of the downstream views created and maintained by hand. Going forward, high-level specification languages can capture this information once and use it to generate and update downstream views. Clearly, applying compilers to hardware isn't a new concept, but applying the machine-readable specification to SoC creation is.
A design team should imagine adding an IP block to its SoC design, configuring its properties, automatically stitching other IP blocks together, automatically running tests based on IP and SoC configuration information, and generating documentation and software headers. Then, anytime the specifications change, a team member would update the information in one place—and only one time. Such correct-by-construction flows will be one of the key areas in 2006.
THE VERIFICATION SIDE
Electronic designers need new methodologies for functional verification. A successful verification methodology encompasses test-bench automation, assertion-based verification, coverage-driven verification, and transaction-level modeling. Critical applications include simulation, formal checkers, emulation and functional prototypes, and the ability to integrate these technologies into a comprehensive verification environment.
It's increasingly important that the verification environment be set up to automatically detect bugs for features such as assertions, automated response checkers, and scoreboards. Furthermore, the environment must generate proper stimulus from directed tests to cause bugs to occur, or from powerful test-benches for constrained random techniques to deploy a wide range of scenarios with a relatively small amount of code. Once test-bench randomization takes place, functional coverage is required to determine which of the possible scenarios actually occurred.
Reprints
