The Monolith Syndrome
Designing and building semiconductors represents the very best of science and mathematics: physics, geometry, thermal mechanics, Boolean logic, complex mathematical modeling, not to mention a good deal of finger-crossing at tapeout.
While much of what has been accomplished is nothing short of amazing, particularly given the time frame in which some of these advances were made, the hardest work is still ahead. The next step isn’t so much creating new dielectrics or layout techniques—although that work will have to continue for Moore’s Law to survive. It’s taking advantage of what’s already been developed in many places and integrating it together into even more complex devices.
This is easier said than done. Just because technologies can co-exist on separate chips or even within the same package doesn’t mean they can be loaded onto the same piece of silicon without creating significant new problems. Heat, which now comes from both static and active power leakage, is one issue that has to be dealt with. Signal integrity is another. Overall reliability is a third. Flexibility, a more recent phenomenon driven by changing profitability requirements, is a fourth.
It’s this flexibility that is raising the most red flags recently. Profitable designs of the future need to be able to span multiple uses and multiple derivatives in each segment in which they are used—different vertical markets, applications that span multiple markets and in various configurations for the same market. And they have to do this without impacting performance or raising the power budget.
As an industry, we are used to creating chips linearly. We have flows, models and business models to match—a monolithic approach to a multi-dimensional problem. The next step will be to re-invent the method while still keeping the rest of the pieces working properly—a combination of the old and the new, and a set of skills to match. The old stuff won’t certainly go away, but the growing opportunity is in integrating it in new ways.
—Ed Sperling
