It’s Transition Time Again

A huge number of changes are needed to get to the next big changes in technology, but who’s going to pay for incremental changes? Answer: Everyone.


By Ed Sperling
After decades of shrinking features, developing new software on every level and engineering huge improvements in energy efficiency and performance, the semiconductor industry has reached a crossroads. To get to the next level will require massive improvements on all fronts, but not all consumers will be willing to pay for them.

For example, if a battery lasts an entire day for a smartphone user, would they be willing to pay more for a couple hours extra battery? Probably not. And how about improvements in voice recognition or slightly better graphics in a video game, or 20% faster performance of an Excel spreadsheet? Again, the answer is highly unlikely.

In most industries, this would be the point where the financial wizards reach for their antacids. In the chip world, it’s the point where other parts of the ecosystem catch up so that the entire ecosystem can move forward. What is now arguably the most complex supply chain in history is suffering from growing pains, although not in the usual way. In this case, not all the parts that comprise the supply chain are advancing at the same pace. And while general improvements are acceptable in upgrades of final products, such as a new phone every couple years, what has become painfully obvious is just how out of sync certain pieces are.

Markets and technology
One change that is unfolding, and showing up the incongruities in the supply chain, is the way computing is being done outside of the office. Computing is no longer confined just to computers, and in the future it will be done by devices talking to other devices.

But we’re used to thinking of a computer as a combination of logic, an interface, I/O and a screen. There is far less focus on the screen going forward, which makes the design process and tradeoffs much easier to understand.

“There’s been a lot of innovation on the hardware side for lower cost and better power management,” said Cary Chin, marketing director for low-power solutions at Synopsys. “What’s changed is what’s a computer versus a display is now a separate issue again. That leaves communication and the power cost for computing, and in many cases traditional computing is moving into the cloud. You can throw in a display of any size, whether it’s a watch screen or a big TV and process the information at any level.”

For example, an iPad can be used as a wireless hotspot without a screen, and a phone may only have to be connected with an earpiece and no screen.

“What this does for low power is huge, because if all the power being consumed is in the display, then who cares about all the other stuff,” Chin said. “But if it’s not just a visual interface, then there’s a lot more that can be done.”

But how much will be done is the big question. It’s one thing to introduce a phone with better connectivity and better resolution in the screen. But will people pay just for better connectivity and longer battery life?

“The answer is no unless there are dramatically new applications,” said Mike Gianfagna, vice president of corporate marketing at Atrenta. “Other than that, you expect cell phones and tablets to cost the same. They could all stand to be improved, and that’s a non-trivial task to improve them. But it’s also the price of admission, at least for the consumer electronics mindset.”

In some cases, doing things better and more elegantly is enough of a change. Gianfagna pointed to a new company called Nest, which employs former engineers from Tesla and Apple, both of which made a fortune from repackaging existing technology in a different business context. They have teamed up to design a more effective thermal management system for buildings based upon schedules of people living or working there, with interfaces ranging from a wall thermostat to a smart phone to sensors communicating with each other without human intervention.

In other cases, it frequently comes down to more rationalized application of resources for a particular function, such as always-on sensors. “The key here is to make it so low power that it doesn’t have to be turned off,” said Eric Dewannain, vice president and general manager of the baseband business unit at Tensilica. “If you do that with an ARM or an Intel processor you have to power them up, which draws a lot of power.”

But just waking up a system with an always-on sensor isn’t so easy. That sensor requires its own power island tied into a signal chain for constant listening and sometimes fast wake-up. Moreover, to make a difference it should draw a maximum of 3 milliwatts of power, and preferably as little as 1 milliwatt, said Dewannain. “Today we’re seeing 3 to 5 milliwatts best case. This kind of technology can go into every appliance, and in the future it will probably have biometric security based upon those voice commands.”

Doing things differently
One area where the individual components are in need of adjustment is on the process side. The fact that companies are now wrestling with choices between finFETs at 20/14nm, stacked die configurations, and fully depleted silicon on insulator with biasing at 28nm, shows just how seriously companies are taking the power-performance-area equation.

There is no single answer for companies in this regard, but process may have repercussions in other areas of the design. “The promise of finFETs is that you don’t need all the other power savings techniques and you don’t have to balance that with performance,” said Mark Murphy, strategic program director at Cadence. “You get X percent improvement right off the top. The challenge is putting the learning right back into the process side so you get PPA right out of the box.”

Another necessary step will be efficiency gains in the manufacturing flows, themselves. Gene Matter, senior applications manager at Docea Power, said the SoC equation is much tougher and will require changes on the back end.

“Someone will come up with a way to take advantage of the new (double and triple) patterning with test and design by combining that with the back-end testing,” said Matter. “We’ve seen that happen before and we will certainly see it again.”

But some of this also will require a different mindset, from the back end all the way to the consumer. “There are kids today who won’t even pay 99 cents to download an application, but they are willing to pay a lot of money to Verizon and AT&T for their phone,” said Synopsys’ Chin. “Some of that money is going back to companies like Apple and Samsung, too, but right now this isn’t transparent. It will be at some point in the future, and then we’ll see what people are really willing to pay for.”