Leibson's Law

Over the years, Ron Collett has collated survey data on more than 1200 IC design projects at more than 35 semiconductor companies. His research results include reuse statistics on some 15,000 IP blocks. Collett delivered some solid conclusions based on that research at the IP 07 conference held in Grenoble earlier this month. The good news is that IP reuse indeed works and delivers enormous and quantifiable benefit.

No one knows more about managing large chip-design projects than Collett. Currently President and CEO of Numetrics Management Systems, Collett spent several years at Dataquest where he oversaw EDA, ASIC, and FPGA research. He founded the research and consulting firm Collett International in 1992, which specialized in developing strategies for semiconductor and EDA companies.

Over the past 10 years, according to Collett’s research, companies have more than doubled their reuse of IP. Some of the many positive, documented results from this increase in reuse are:

• A linear reduction in project effort (measured in person-weeks)
• A linear reduction in project duration (resulting in faster time to market)
• A linear reduction in the number of silicon spins needed to obtain a functional design
• A linear reduction in schedule slip (measured as a percentage of overall schedule time)

These IP reuse benefits have accrued despite an exponential increase in IC complexity (thank you, Moore’s Law). However, the benefits didn’t come for free. Team sizes have doubled, just since the year 2000 and 85% of all IC design projects still slip their schedules. One continuing problem is that not all blocks are equally reusable. I asked Collett to tell me the most reusable type of IP block. “Memory” was his immediate reply.

IP blocks like memory are essentially 100% reusable. There’s tremendous design productivity to be gained by using such blocks. Other blocks cannot be fully reused for a variety of reasons. Perhaps the function isn’t exactly right for a new design and the block must be altered accordingly. Perhaps the block lacks adequate documentation and its functional specifications must be recovered through reverse engineering before reuse is possible.

During my discussion with Collett, I posited that most microprocessor IP also falls into the highly reusable category. Surprisingly, Collett questioned my premise. I replied that microprocessor cores are easily reused because they’re some of the most well-documented IP cores available. There are usually big, fat ISA (instruction set architecture), user, hardware designer, and programmer manuals plus sample code, application notes, software tools (compiler, assembler, linker, debugger, instruction-set simulator), synthesis scripts, and simulation models that accompany the microprocessor core’s hardware block. In addition, a microprocessor’s function can be changed through firmware, without altering the hardware IP.

A moment’s thought about those points and Collett agreed. (Hey, with my EDN, Microprocessor Report, and Tensilica background, I just had to ask. I’ve got a microprocessor-centric brain.) Collett’s thesis is that unless the IP block can be fully or nearly fully resused, the benefits of reuse don’t kick in and the benefit function is highly nonlinear. So why is Collett on this particular IP-reuse soap box? Numetrics, helps companies manage IC-design project risk and ride herd on project schedules through consulting and by offering software that gives project managers access to the industry norms derived from the company’s large project-history database. Managers can see if their proposed project schedule, productivity, and staffing estimates are in line with industry norms or if their schedules are irrationally exuberant.

For example, if the projected schedule requires the design team to be three times more productive than industry norms, it may be overly optimistic. However, if the team consists of a battle-hardened group of rocket scientists, the schedule might be realistic. Even if the design team doesn’t consist solely of superstars, a plan to push IP reuse to the max might still bring the schedule into a more realistic realm. Numetrics also provides consulting along these lines.