Poor schedule predictability of IC development projects is the Achilles heel of semiconductor companies. It manifests itself as high schedule slip and is among the most important R&D metrics, measuring how well project schedules reflect reality. Most don't.
Companies traditionally view schedule slip not as a result of faulty project plans, but rather as a consequence of unforeseeable perturbations occurring during the development process. The picture is incomplete and inaccurate. Slip must also be viewed through the project planning lens, because many events labeled as unforeseeable can be fully contemplated in the project plan with proper modeling. The payoff is big—reliable plans, which is the path to profitability.
Predictability and schedule slip are two sides of the same coin. Schedule predictability measures schedule overrun by comparing a project's original planned duration to its actual duration, expressing the difference as a percentage of the original. The original planned schedule is the first one formally defined and officially disseminated.
What qualifies as low versus high slip in the semiconductor industry? Low is 10 percent or less. High is more than 25 percent.
Poor schedule predictability is an insidious problem that eats away at a company's competitiveness, especially when it becomes fully baked into the organization's culture as a tolerated practice. At that point, it's just a matter of time before the company or business line disintegrates. Today many semiconductor organizations are taking action to vanquish slip. Those ignoring it do so at their peril. I doubt they'll be around five years from now.
Excuses abound during post mortem assessments of projects that slip schedule. Naturally, the usual culprits are speciously defended as wholly unforeseeable. They include spec changes, EDA tool and library problems, IP and software delays, personnel issues, etc. But what project doesn't encounter some or perhaps all of these and more? It's disingenuous to expect that any particular project will escape the stochastic nature of chip development, so therefore it must be built into the plan—without the use of schedule buffers at every turn.
I reject the notion that IC projects are predictably unpredictable. I can point to myriad organizations that have excellent predictability. These groups deploy best practices and tools that use facts and data to ensure project plans fully contemplate the stochastic nature of IC development and "unforeseeable" events.
Management thrives on data -- it makes decision-making so much easier, because it takes some of the risk out of the equation. Hard facts and data are therefore a powerful tool for persuading management that a project's schedule is wholly unrealistic. The data must unequivocally show that even if the team acheived best-in-class productivity, the schedule is still not achievable because the staffing allocated to the project is not communsurate with the design's complexity and the schedule constraint.
When presented with this, management has little choice but to reduce the scope of the project, relax the schedule constraint, add more resources and/or reblance the project portfolio (i.e. reduce the number of projects).
Thanks for the comment.
An interesting article, I have worked in the semi industry for more than 15 years and never found a schedule realistic. Too often the designers presented a real schedule to the management and it was tweaked to meet some arbitrary set of deadlines. This is oftentimes the real cause of "schedule slip" it is non-realistic schedules being adopted as real. Specs do change, resources do disappear, software has issues these are not unexpected and should be accounted for but as I stated the real schedule does not meet the management's target dates. Until that is addressed, development will always be "late".
Well stated -- it is all about company culture. Politics and project planning often go hand-in-hand, because the project planning phase is when funding decisions are made. It is almost axiomatic that money begets politics. In my experience, the most effective and efficient way to cut through politics is with facts/data, which incidentally has a tremendously positive impact on culture. Once established, a fact/data-driven culture can persist for a long time, which is a key ingredient for sustaining competitive advantage.
Thanks for the comment.
It is definitely obvious -- but remarkably, a great many semiconductor organizations don't understand it. As for an advertisement, you are again correct -- it is an advertisement to think differently about schedule predictability.
Great article. CPM schedules are inherently challenging as the forecasting tool of choice for development type projects. Due to the high degree of uncertainty and iterative work/re-work, traditional CPM Gantt chart type forecasting rarely gives an accurate picture - hence the frequent perception that the project finished late - i would suggest that instead, we simply didn't plan accurately enough. Worse, given projects teams are aware of this shortcoming, the original 'plan on the wall' simply becomes a static picture that doesn't reflect the latest updates, changes and impacts due to risk. An alternate approach is to create a risk-adjusted schedule using a combination of a CPM schedule (e.g. MS Project) combined with a project's risk register and model the impact of the latter on the former using Monte Carlo analysis. Not only does this approach give more realistic forecasts, it is also a means of getting team buy-in into the agreed upon plan. I agree with the author that these projects are indeed predictable; we simply need to do a better job of including the non-deterministic work/outcomes in the original plan/schedule. More thoughts on this at http://goo.gl/KQTuz
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.