In the move to stacked die, one of the biggest issues is power. While chips using leading-edge processes already address this issue effectively through a variety of advanced techniques, the big question mark is what happens with the older technologies.

The answer may not be quite so simple. While it’s still possible to use technology developed at older process nodes, it may not be exactly the same technology as was originally developed at those older nodes. In some cases, it will be completely different. Much of the IP that will be “re-used” will likely be IP that was developed specifically for stacked die, not for older chips.

The key advantage is that the process technology has been tested, worked out, and modified appropriately for 3D, not that the IP design won’t need to be tweaked. In the analog world, this may still amount to a significant redesign of that IP.

The big culprit here is power, and power budgets on stacked die won’t be any more lenient than on two-dimensional structures. In fact, they may be more constraining because the die will be thinner. In 2.5D stacks it will be easier to use thicker die, but the power constraints will likely be similar. Battery technology isn’t changing quickly enough, and more features with better graphics and faster performance are likely to absorb most of the gains of approaches such as Wide I/O.

The result is that stacking of die won’t get easier. And while there will be time savings and yield improvements once this packaging approach ramps to commercial viability, it will take time to get the whole ecosystem updated and aligned. When it comes to SoCs, there still are no shortcuts, and probably won’t be for years to come.

–Ed Sperling

A mad scramble is underway—quietly, of course, because it involves competitive positioning—to prepare for stacked die. No one wants to be left out of this transition, regardless of whether they’ve branded it as pure fiction in the past. And no one wants to be caught unprepared, even though they’re not sure exactly when this shift will take place.

But what’s particularly interesting underneath is exactly how companies are jockeying for position. Rather than working at the latest technology node, which is where most battles have been fought in the past, stacked die involves lots of older process nodes as well as the newest stuff. As a result, the race is happening in multiple areas—at the bleeding edge of technology and as far back as 250nm. All process technology is being retrofitted with new techniques for manufacturability, yield, test, and power. In fact, what is happening at the forefront of Moore’s Law is now being pushed backward instead of just the reverse.

The manufacturability and yield are relatively well established, even though they are being improved. Test may actually require a different sizing of die and some adjustments in layout and packaging. But the really big change is on the power side. Before 90nm, power was an afterthought. Computers were still the driving force for semiconductors and most of them were built for speed and plugged into a wall. Two hours of battery time was considered acceptable. Shutting down parts of a chips was considered ridiculous.

Fast forward to the present and energy is now a critical factor in everything from mobile electronics to data centers. That won’t change with stacked die, either. Power is an essential ingredient for success. Gaining efficiency or performance with Wide I/O while losing it with older analog or digital components in a multi-layer stack is a recipe for disaster, and changes are under way to make sure that doesn’t happen. The only question now is how quickly.

–Ed Sperling

The focus in recent mobile designs has been heavily slanted toward energy efficiency and power reduction, but this is merely a swing of the pendulum. The loudest complaint right now is time between battery charges, and devices that can go longer between charges without having to turn down the screen brightness, turn off Bluetooth and limit the number of e-mail refreshes will win market share.

While time between charges will remain important, there are massive efforts underway to improve everything from how radios work to reducing the distance between memory and logic—and even reducing the amount of energy needed to drive bits back and forth between logic and memory. Once these efforts bear fruit, probably over the next couple years, the pendulum will begin swinging back toward the center. Performance and efficiency will be the tradeoffs, and they will depend upon user preferences more than default factory settings.

The trick will be to make these kinds of changes more seamless, rather than requiring users to adjust the settings within their handheld devices. Programming a smart phone or a portable medical device or an infotainment system on a car isn’t something the average user wants to spend time learning. That’s why VCR makers ultimately decided that the best way to deal with the blinking 12:00 was to remove the clocks entirely, and it’s why BMW has had to modify the iDrive controller.

What’s needed isn’t better segmentation. It’s more intelligence built into the device itself to say what applications are being run, how bright the screen needs to be and how much energy is required. A user who is playing a game or downloading an HD movie wants maximum power, while someone doing e-mail may be able to limit their power consumption dramatically—particularly if they’re not in direct sunlight.

The tradeoffs of low power and performance are at the center of these use models, and it will be up to designers—and tools makers—to begin to integrate these approaches much more intelligently. In the near future it will no longer just about the number of power islands and voltage rails. It will be about being able to more easily move from the most efficient to the most performance without having to fiddle with device programming.

For the user, this is all about simplicity. For the architects, design engineers and the developers of the automation tools they use, this is about making things simpler for the end user. And as everyone involved knows, that’s hardly a simple task.

–Ed Sperling

The human memory is rather short when it comes to certain things. Energy efficiency is one of them. While they may cringe at paying $4 a gallon for gasoline to fill of their car, they were convinced that drastic measures were necessary when gas hit $1 a gallon.

And while consumers collectively account for the vast percentage of energy consumed, individually they don’t consume enough to make the kinds of hard choices that have been predicted for the better part of two decades. Solar is still not a high-enough volume market to make it attractive without government subsidies. Hybrid vehicles still aren’t economically beneficial, and all-electric vehicles might never hit volume without some serious subsidization.

Businesses—particularly large corporations—are another matter. One of the basic reasons for existence of corporations is that they can achieve economies of scale on all fronts. That also means they have a reason to act on things that to most people are just an annoyance, such as rising energy costs. While an increase of $100 a year may not be enough motivation for a homeowner to replace a computer or even a water heater, an increase that runs into the six or seven figures can get very quick action inside a company.

Banks and credit card companies have enough money in play to even see a significant gain from a float, the interest gained by delays in paying their own bills over a short period of time. That’s sometimes even noticeable in changes of a day or two. When airlines cut out pillows they save millions of dollars. And when employees take their portable devices home at night and charge them they’re probably contributing significantly to their employer’s bottom line. Multiple even a dime a day for 1 KwH by 100,000 employees and it comes out to $50,000 a week, or $2.6 million a year.

The same will be true of cloud-based services and shared resources of any sort. The challenge has always been maximizing the use of resources—idle time costs money—while making sure enough resources are available to do the job. In the EDA world there will be a clear future for cloud-based services such as emulation, debug and verification, in EDA, just as there are clear advantages to centralized computing in other areas. Apple’s new operating environment, Lion, is a recognition of the importance of shared resources. So are most of the new services being offered by companies such as IBM, Microsoft, Amazon, Oracle and HP.

In consumer electronics, it’s not so much economics driving the shift to more efficiency as the annoyance of shorter battery life. And in cars, the big question now is whether time between fill-ups is more important than a few extra miles per gallon or per charge. But for any technology makers looking to really earn profits on green, the key will continue to be big bottom-line savings among companies and organizations that have sufficient scale to make those savings really matter.

–Ed Sperling