The future of SoCs: software as differentiator and demon

Speakers at Semico’s Outlook 2010 briefing in San Jose yesterday painted a detailed image of a world in which ubiquitous microelectronics took on a host of troubles—from a warming, energy-starved globe to an aging population—and triumphed over them, in the process finding a path to continued growth in an uncertain world economy. But lurking in the corner during much of the discussion was an issue not widely discussed in chip-design circles, yet increasingly critical to this happy scenario. That issue was software.

Teridian Semiconductor vice president and general manager Jay Cormier, for example, described the diffusion of Smart Grid electronics into the power network in four successive waves. First, Cormier said, SoCs that monitor a device’s state, communicate with the grid, and manage the device’s power have begun to appear in the areas of greatest potential savings: giant data centers. The second wave will carry the technology into smart buildings, spawning HVAC, lighting, and office-equipment advances. Then two more waves will roll over consumers: first, grid-friendly white goods will hit the market, and in the end we will begin to see full-home automation.

Similarly, NXP senior vice president of sales and marketing Michael Noonen foresaw a buildout of microelectronics into four broad application areas: energy, healthcare, mobility, and security. And IBM fellow and chief Power Systems architect Jim Kahle traced the same vector from another point of view, postulating a world filled with perhaps a trillion sensors, the networks to concentrate their data, and the processors to fuse that data into a real-time model of the world and to make command-and-control decisions. Kahle offered as an early example the city of Stockholm’s  system for measuring traffic density and assessing congestion-based tolls using optical recognition of license plates.

In a later presentation, Freescale senior vice president and general manager Lisa Su sketched an SoC architectural evolution similar to what Kahle had described, but with the objective of creating a single-chip base station. Su told how SoCs had moved from processor-centric chips to symmetric multicore systems, and now on to heterogeneous clusters of general-purpose processors and function-specific accelerators.

In all of these visions the shadow across the back of the stage was software. "Exploiting this level of integration," Su warned, "will require a great deal of software optimization."

Bradley Howe, vice president of IC engineering at Altera, linked the importance of software directly to chip vendors. "As far as the system developers are concerned, the semiconductor vendor owns the efficiency of the algorithms," he said. Other speakers reiterated the point, saying that today an SoC is not a product without a full suite of application-specific software to support it. And Noonen took the idea a step further. "We are seeing a shift in the source of differentiation: from hardware to software," he stated.

And yet hardware still has a key role, if increasingly a supporting one. An audience question asked if software development flows were really up to the task being set for them, given recent questions about software’s responsibility for some high-profile product failures.

"Software is a serious challenge," agreed Noonen. And Kahle added that sometimes hardware had to share that challenge—making architectural and implementation choices specifically to ease the burden on software development. "You have to use hardware to drive simplicity into the software," he said.

But during an afternoon keynote, Mentor Graphics chairman and CEO Walden Rhines said that in the end, the scale of the problem would require software development to change as well. Rhines began his argument with the constantly-repeated mantra that IC development cost is spiraling out of control. Then, showing data to support his point, Rhines said "In reality, the cost of IC hardware design is not increasing—it has held fairly steady across recent process nodes. It’s the cost of the software that supports the chip that is growing."

Rhines said there was no choice but to control software costs, and that this would happen. "I believe it will be controlled, but it will take a number of measures," he said. "We will see standardized, application-specific middleware such as Android. We will see increased design reuse, as we have in hardware, and the emergence of more open standards. And we will see increased design automation, including software high-level synthesis environments like Autosar and xtUML."

The recognition of the centrality and scale of the software problem so far does not seem to be accompanied by a proportional direction of resources toward solving it. But as the industry begins to ponder the fact that a task outside our normal definition of SoC design is both determining the differentiation of our products and consuming the bulk of our development expense and schedule, expect that to change. No, insist that it change.