Compression Effect

For all the talk of restricted design and increasing complexity, it seems that what’s really happening is that restrictions are being lifted off one group and placed on another.

This happens from time to time in system-level design, usually at an inflection point in the overall system development process or at the start of a new process node. For the past couple decades, much of the creativity in semiconductor development was done at the design engineer level, often but not always in conjunction with the chip architect, while most of the heavy lifting and saving the day was done at the back end in verification and in manufacturing.

At 32nm and beyond, it’s the design engineers who are losing some freedom to say how things can be placed on a chip, what can be done to fix design flaws after the fact and what new approaches can be added. Architects now have more control over the whole SoC than ever before—and there are plenty of people who will argue the pros and cons of that for years to come—and the foundries are staking a claim from the back-end to make sure that what’s designed can be manufactured. Nobody makes money if there are yield problems or if a chip is late to market.

What’s driving this shift are too many variables on the front end and too much data on the back end. Architects have to wrestle with power budgets, power islands, business issues and marketing priorities. Foundries have to deal with massive amounts of data that get even bigger with computational scaling, and verification and debugging still have to deal with all of these issues. Costs are rising, respins can force companies to lose market windows, and complexity is forcing some choices that didn’t have to be made a couple process nodes ago. And in the midst of all of this, companies are now responsible for developing larger and larger portions of the software stack.

In between are the engineers who work at various phases of the flow, from design to layout to verification, watching what appears to be a massive shift in their market on all sides. They’re correct in sensing the changes. These are significant, and it’s too early to tell which jobs will ultimately rule the roost or at which process nodes. But whatever changes occur, you can be certain they won’t be the last.

—Ed Sperling