At the recent International Solid-State Circuits Conference, Jay Heeb, Intel architecture manager, raised more than a few eyebrows by stating that the SoC movement is "dead." Heeb, who manages Intel's X-scale processor core design effort, made his comments at an ISSCC session on the future of ICs for 3G mobile phones.

SoCs require too many mask layers, Heeb said. A better option is what he called a "system-on-3D" package, distantly related to today's stacked packages. Heeb said that by the end of the decade, 90 percent of the transistors in a cell phone's chip set will be nonvolatile memory. Rather than integrating all that on a single die, he said, it will be better to create a system with 3-D integration capabilities.

To support 3-D packages, Heeb said, tools will be needed for 3-D floor planning, routing, estimation, extraction and debugging.

This speech reminded me of a talk I had recently with Mark Rencher, the president of Pivotal Enterprises, a consulting firm that helps users enable SIP/MCM design flows. Despite all the hype about SoCs, Rencher said, many RF platforms are implemented on MCMs, including seven out of 10 Bluetooth designs. MCMs are also showing up in processor design, medical electronics and automotive applications, he said.

"It's a good, viable solution for integrating chip sets into a subsystem or platform," he said. "You can mix technologies on the MCM. You can use less expensive technologies to achieve the same results and, if you have to do manufacturing debugging, it's much faster than SoCs and much cheaper."

But there's a glaring problem, Rencher said, and that's the lack of EDA support. "None of the big vendors are saying they're going to support a solution for MCMs," he said.

What's needed, Rencher said, is design capture that supports RF, layout tools with Gerber manufacturing links and electromagnetic-analysis tools. Traditional pcb tools don't cut it, he noted, because of factors like resistance and capacitance in transmission lines.

There are other alternatives to SoCs, such as FPGAs and programmable chips from companies like PicoChip and QuickSilver. The EDA industry is making a mistake if it puts all its eggs in one basket.