Facts and data versus heuristics and hope
11/14/2010 10:45 PM EST
During the next five years, a great many semiconductor companies will be faced with an increasing number of underperforming business units. Chances are they'll be selling or spinning them off. Some chip companies, large and small, will disappear altogether. Why?
Persistently missed product development schedules—poor schedule predictability—is the culprit and bane of semiconductor companies. It is a pervasive problem in an industry whose R&D stakes are now excruciatingly high. Typical SoC development costs, for example, range from $50 million to $100 million (from design concept to release-to-production). With a target return of 5X to 10X and a narrow market window, a mere three-month slip can dramatically reduce revenue and profitability. No surprise.
Projects miss schedule when management underestimates or fails to acknowledge the time and resources the R&D organization needs to develop complex ICs. Absence of a reliable estimation process is the crux of the problem. Curiously, many semiconductor executives have been blind to the issue, and yet it is the key failure mechanism of their businesses because it goes to the heart of their product development engine. Go figure.
Accurately calculating design difficulty—the prerequisite to reliably estimating resources and schedules—demands quantitative methods that measure the intrinsic difficulty of designing an ICs logic, circuitry, packaging, etc. However, it must also take into account the sizable stochastic footprint of chip development.
Some executives fail to recognize the stochastic aspect of IC development has a far greater impact than ever before—a consequence of soaring development costs, inexorable competition and the limited size of each market opportunity. Resource and schedule planning therefore demands a stochastic model, which contemplates events that projects routinely encounter. Examples include spec changes, EDA tool issues, IP quality, project management, organizational issues, etc. It's not exceptionally hard to model the development process, but it does require a good deal of industry data.
Project and program managers typically rely on experience and intuition to create project plans. Not a bad start, but the approach is light on facts and data and heavy on heuristics and hope. IC development has both a deterministic and stochastic component, and they are inextricably intertwined. Most project plans contemplate only the deterministic aspect of complexity, which is why they are flawed from the start. Not addressing the "stochastic problem" will almost certainly translate into reduced shareholder value and lost jobs.
Ronald Collett is president and CEO of Numetrics Management Systems, Inc. www.numetrics.com