Which Comes First?

Methodologies in IC design typically follow tools. The tools enable the methodologies, and chipmakers’ businesses are built around both of them. That has been the rock-solid foundation for the design and production of chips since well before the impenetrable 1-micron wall.

But that approach is falling apart at 28nm, and it will continue to crumble at 16/14nm and 10nm. It simply isn’t fast enough to continue a serial flow, where one operation precedes another. Those operations—architecture, software development, floor planning, synthesis, verification, and even collaboration on process changes with the foundry—now have overlaps. Some of those overlaps are bigger than others—software prototyping and RTL development, for example—but increasingly the design flow looks like a complex matrix rather than a series of connected vectors.

EDA tools, for the most part, have handled the increased amount of data processing and the cross-function communication rather well. Businesses processes are another story. Because chips now take longer to produce than market windows allow, especially those that involve moving from one node to the next every couple years, they now have to be started in parallel. That also means all of the operations involved in developing those chips have to be updated continually, because many of these chips are started well before manufacturing processes reach version 1.0. At advanced nodes, even version 1.0 isn’t good enough for widespread production, although it’s a lot better than where many of these designs are initiated.

This requires much more agility than many chipmakers have built into their internal business processes. The big chipmakers are fully cognizant of this fact, and they are making changes accordingly. Midsize chipmakers are struggling. What very few of them are doing, however, is thinking about what they should be doing differently from a business standpoint. For example, are they using the best methodologies for verification and will that make a difference? Are they cross-training hardware and software engineering teams? Are they building bridges and training people internally to be able to span multiple departments and shift from one narrow area of expertise to what has been rather vaguely described in the past as systems engineering?

There are other considerations, as well. While many companies understand the platform approach quite well, there is far less knowledge about exactly who should create the various pieces. This has a direct bearing on the most time-consuming area in design these days—software—because the more black-box IP that is used, the more standardized interfaces will be required. This is good and bad. It saves time, but it also requires a detailed knowledge of how that software works with IP blocks in order to improve performance and, more importantly, reduce power.

So far, most of the standards for IP involve integration, not power. New standards need to be developed, and at this point it’s questionable whether there is enough collective knowledge in the industry to make that happen on a grand scale.

And finally, businesses need to reassess whether they will continue to push everything into finer and finer geometries, or whether they will begin piecing together platforms and how those platforms will be packaged. Work has begun in this area. Even Intel, the torchbearer for Moore’s Law, has moved memory off chip and created a low-cost connector (Embedded Multi-die Interconnect Bridge) that can be used in place of an interposer. What kind of effect will 2.5D and 3D-ICs have on business processes? Can companies adapt quickly enough, particularly when they’ve just focused their resources and attention on getting 28nm, 16/14nm, and 10nm chips out the door in parallel?

Agility in the business of technology will become increasingly important for survival rather than just the ability to create new chips. Moore’s Law, for all the direction it has provided for technology, was at its heart a business statement. As it becomes increasingly difficult to sustain, it will be the business side that will have to contribute heavily to finding new directions, not just technology breakthroughs.